# Your Agent's Biggest Lie: "I Searched the Web"

Rafael Levi

AI Engineer Europe 2026

[Talk page](https://ai.engineer/talks/btxGmN8RvNU-your-agents-biggest-lie-i-searched-web) · [Video](https://www.youtube.com/watch?v=btxGmN8RvNU)

## Your Agent's Biggest Lie: "I Searched the Web"

Reliable web agents need more than plausible answers: they need usable retrieval tools, explicit failure reporting, and a clear distinction between accessing a page and verifying its contents.

Familiarity with LLM tool calls and basic web requests will help; MCP is the protocol used here to expose external retrieval tools to the agent.

### What evidence is behind the answer?

[0:19](https://www.youtube.com/watch?v=btxGmN8RvNU&t=19s)

When an agent says it checked the web, what establishes that it actually obtained usable information? Rafael Levi, introducing himself as representing Bright Data, starts with the gap between an agent’s confident account of its work and the access it really has. A request can encounter a CAPTCHA or another block before it reaches the data. Levi argues that a model’s tendency to satisfy the user can then hide the missing evidence behind a plausible answer.

![Slide stating that AI agents get blocked, receive fake content, and hit CAPTCHAs, with the speaker inset below.](https://ai.engineer/api/talk-essay-media/255d96d625755688259a64cde5f6160720ead1cb673b7a3d9e3b0d06f631ba6c)

Agents encounter blocks, fake content, and CAPTCHAs, then report back as if nothing went wrong.

The dangerous substitution is **training knowledge presented as live information**. Levi illustrates this with information from 2024 being offered as current in 2026; that is his example of stale knowledge, not a statement about every model’s cutoff. He estimates that Cloudflare makes roughly 20% of the web inaccessible to default AI fetching, but supplies no measurement establishing that access rate. Cloudflare’s [crawl-policy announcement](https://blog.cloudflare.com/content-independence-day-no-ai-crawl-without-compensation/) does not establish universal blocking across that share of the web.

There is also a failure mode in which a crawler receives content that is irrelevant to the intended source. Levi describes [Cloudflare’s AI Labyrinth](https://blog.cloudflare.com/ai-labyrinth/) as feeding bots fake data. Cloudflare’s documented mechanism is more specific: an opt-in system introduces hidden links that lead suspected unauthorized crawlers into generated decoy pages. Those pages contain unrelated scientific facts, rather than deliberately false facts. Either way, obtaining text does not establish that the text answers the original request.

The resulting failure can be invisible: the request returns a CAPTCHA or an empty page, but the answer contains neither an error nor a warning. Levi describes fabricated numbers and citations that lead to nonexistent pages. He estimates that roughly 60% of ChatGPT citations do not work, without giving a sample or methodology. That is not an established general broken-link rate. His concrete example is a shopping request: the agent recommends a product for $50, but clicking the supplied link reveals neither a valid destination nor a product to buy. **A plausible product, price, and citation are not evidence of a retrieved listing.**

### Five URLs, with no browsing tools

[3:01](https://www.youtube.com/watch?v=btxGmN8RvNU&t=181s)

The demonstration switches to VS Code and compares two scripts with identical prompts: one without the [Bright Data MCP server](https://github.com/brightdata/brightdata-mcp), and one with it. The tasks target Rightmove properties, Mistral’s LinkedIn company page, an Instagram account, an Amazon product, and TikTok. These are supplied destinations, selected because Levi expects their anti-bot systems to make access difficult.

The first run uses **GPT-5** without browsing tools. It explicitly reports that it has no live web access. The tool-free baseline records zero successes and five failures across the selected URL tasks. This is a capability comparison, not a representative benchmark: Levi deliberately chose these sites, and the two runs have different tools despite receiving the same prompts. The baseline also admits its limitation, so this particular run does not demonstrate the concealed failure described in the opening.

![VS Code showing URL tasks above terminal output with live-access limitations and a scorecard of 0 succeeded, 5 failed.](https://ai.engineer/api/talk-essay-media/5fcb03a10430fbd13d7d87c4a2aae1656f4d85e3f4b61bae835415f9158dce6b)

The tool-free baseline finishes with zero successes and five failures.

### What the MCP connection adds

[4:52](https://www.youtube.com/watch?v=btxGmN8RvNU&t=292s)

Levi starts the second run with an MCP server exposing 69 tools. The first capability he describes is search: the agent can issue queries to Google, Bing, or DuckDuckGo rather than merely produce an answer that sounds like a search result. Next comes **Markdown extraction**, which returns page content without HTML tags so the model spends less context on markup. Batch search extends the query operation to multiple keywords; Levi illustrates it with 100 keywords. These are descriptions of the demonstrated tool surface: the current repository instead lists Google, Bing, and Yandex and documents batches of at most 10 items per call.

The remaining capabilities cover two different ways to obtain data: prebuilt APIs for supported sites and a remote browser that the agent can navigate. Levi describes browser infrastructure that solves CAPTCHAs automatically, assigns distinct fingerprints, and supports parallel sessions. He uses 100 simultaneous browsers as an illustration, not a measured concurrency result. The useful distinctions are:

| Capability | Operation | Purpose |
| --- | --- | --- |
| Site API | Return site-specific data | Use a prepared extraction interface |
| Remote browser | Open and navigate pages | Handle interactive access |
| CAPTCHA handling | Resolve an access interruption | Let navigation continue |
| Parallel sessions | Run multiple browser sessions | Collect from several sessions concurrently |

These functions sit beneath the model’s reasoning: they determine what evidence the agent can obtain before it writes an answer.

As results arrive, Levi identifies successful retrieval for Rightmove, Mistral’s LinkedIn page, Instagram, and the Amazon product. He does not explicitly narrate a TikTok result at this point. He then asks the LLM to compare the two runs. That comparison is another output of the demonstration, rather than an independent check of the returned data.

### Public visibility is an access boundary

[6:52](https://www.youtube.com/watch?v=btxGmN8RvNU&t=412s)

An audience question makes the social-profile workflow more concrete: does the service operate accounts to retrieve those pages? Levi says it works only with publicly available data. He connects that policy to the terms accepted when creating an account and to litigation over scraping. This is the service boundary he describes, not a universal rule that authenticated collection is illegal or public collection is legally risk-free. [Privacy authorities’ guidance on scraping](https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2024/10/global-privacy-authorities-issue-follow-up-joint-statement-on-data-scraping-after-industry-engagement/) explicitly applies privacy obligations to publicly accessible personal information.

When the audience questions how much LinkedIn or Instagram exposes publicly, Levi opens a LinkedIn URL in an incognito window and shows that some information is visible without signing in. The discussion then moves from company pages to people’s profiles. He attributes access differences to IP reputation and request patterns, including the shared network at an event. He suggests that a home connection might allow access to 5–10 profiles before prompting for login; this is an anecdotal, conditional estimate, not a reliable quota.

The follow-up question is whether users can supply credentials for their own social accounts. Levi’s answer remains no: the service described does not accept credentials to collect data behind login. An unauthenticated page view and an authenticated account workflow therefore belong on different sides of this integration’s stated boundary.

### Choose freshness, then inspect the result

[8:43](https://www.youtube.com/watch?v=btxGmN8RvNU&t=523s)

A question about caching introduces an alternative to live collection: query an existing dataset. Levi’s example is finding LinkedIn AI engineers in a particular area. If information that may be a few months old is acceptable, the agent can filter stored records and return the matching subset instead of visiting each profile again.

| Route | Operation | Freshness tradeoff |
| --- | --- | --- |
| Live collection | Retrieve the source now | Requires current access |
| Existing dataset | Filter stored records | May contain months-old information |

The answer describes a dataset product, not a cache invalidation policy or a promised refresh interval. The application still needs to decide how current its answer must be.

Returning to the terminal comparison, Levi reads failure and success labels for the two runs, with the baseline characterized as lacking live web access. The spoken readout does not clearly map every label to a site. He then explains the browser’s intended continuation behavior: when navigation encounters a CAPTCHA, the browser handles it internally so the agent can resume browsing. That removes an access interruption; it does not itself validate the information eventually returned.

### Avoiding a block does not verify the content

[9:51](https://www.youtube.com/watch?v=btxGmN8RvNU&t=591s)

Levi’s proposed remedy is to improve retrieval access through MCP so the agent has less reason to fill missing evidence with an unsupported answer. He returns to both blocked requests and misleading content, then introduces a free request allowance for trying the integration. The mechanism addresses an important prerequisite for grounded answers—obtaining source material—but successful access alone cannot establish factual correctness.

An audience member asks how the system detects AI Labyrinth. The answer is an avoidance strategy, rather than a demonstrated content detector: Levi describes prerecorded mouse movement and human-like typing intended to keep browser sessions from being classified as bots. No procedure is shown for recognizing a decoy page after it has been retrieved.

The difficulty becomes clearer in his hotel-price example. A phone may receive one price, a desktop another, and a proxy connection a third. Those differences do not establish which price is correct—or whether the cause is deception, personalization, location, or another access condition. Levi’s proposed response is to make the agent look human and, in his words, “hope for the best.” That qualification matters: the approach tries to avoid a problematic response, but does not prove the response’s validity.

Levi acknowledges that he has little statistical evidence about Labyrinth and describes it as having launched about a month earlier. Its announcement was actually March 19, 2025, more than a year before the [April 2026 event](https://ai.engineer/europe/2026). He reports no observed degradation while collecting petabytes of data daily and comparing cached results. The talk supplies no measurement protocol or independent validation for that observation, so it cannot establish immunity to misleading content.

### Have the model build the parser

[12:26](https://www.youtube.com/watch?v=btxGmN8RvNU&t=746s)

The discussion then shifts from access to the cost of processing retrieved pages. Levi points to GitHub and [Bright Data Agent Skills](https://github.com/brightdata/skills), describing instructions and API information that an agent can use to build a scraper or collection pipeline. He previews a later session in which an agent will build a Walmart collector. That pipeline is promised for the later session, not built in this recording.

The architectural change is to **use the LLM to build a reusable parser**, then run that parser in ordinary code. For a workload such as 10,000 pages, the model need not interpret every page’s HTML separately:

1. Have the agent develop the extraction logic for the target page structure.
2. Put that logic in a parser that a script can invoke.
3. Run the script over the collected pages.

Levi claims approximately 99% token savings for the proposed reusable-parser approach, without a measured baseline or test details. The useful principle does not depend on that percentage: repeated extraction moves out of model calls and into reusable code.

### Expose only the tools the task needs

[13:41](https://www.youtube.com/watch?v=btxGmN8RvNU&t=821s)

The next performance question concerns the 69-tool surface: must an application load every tool just to search? Levi says no. The broad list was there to demonstrate capability; for simple retrieval, he would expose only search and Markdown scraping. **Tool definitions consume context**, so irrelevant capabilities add material the model must process without helping the task.

An application can express that boundary as an allowlist over the tool definitions returned by its server. The TypeScript below illustrates the filtering operation; the caller supplies the actual search and Markdown tool names from its chosen server version:

```typescript
type NamedTool = { name: string };

function selectRetrievalTools<T extends NamedTool>(
  availableTools: readonly T[],
  searchToolName: string,
  markdownToolName: string,
): T[] {
  const allowedNames = new Set([
    searchToolName,
    markdownToolName,
  ]);

  for (const name of allowedNames) {
    if (!availableTools.some((tool) => tool.name === name)) {
      throw new Error(`Required retrieval tool is unavailable: ${name}`);
    }
  }

  return availableTools.filter((tool) => allowedNames.has(tool.name));
}
```

Pass the returned definitions to the model instead of the full list. Filtering preserves the selected tools’ definitions while excluding unrelated ones; it does not change their underlying retrieval behavior.

### The experiment tested URL access, not web search

[14:07](https://www.youtube.com/watch?v=btxGmN8RvNU&t=847s)

The final technical question sharpens the interpretation of the experiment. Did it use OpenAI’s web-search tool, or only call Bright Data? Levi clarifies that the task was to load supplied URLs: no searches were performed. The completed terminal comparison visible at 14:36.740 labels Demo 1 as 0/5 succeeded and Demo 2 as 5/5 succeeded. Those are the displayed LLM-generated verdicts for this selected URL-loading demonstration, not independently verified retrieval scores.

![VS Code showing five URL-based tasks and terminal results with a final verdict of Demo 1: 0/5 succeeded and Demo 2: 5/5 succeeded.](https://ai.engineer/api/talk-essay-media/5fa5c8d953ed39b7183a189ced253c5e99df0f4334f6362cd5d042105dd97b06)

The editor displays supplied-URL tasks above a terminal comparison ending with zero of five and five of five successes.

Levi reiterates that no search tool was used, although Google search is available through the MCP server. That separates three capabilities that an agent’s answer can easily blur:

| Capability | What it establishes |
| --- | --- |
| Generate an answer | The model can produce text |
| Load a supplied URL | A tool can attempt a known destination |
| Search by query | A tool can discover candidate destinations |

The comparison demonstrates the difference between no browsing tools and an enabled retrieval integration. It does not compare Bright Data with OpenAI’s web-search tool or establish which search backend an agent uses merely because the user expects Google-like results. Levi invites the audience to try the integration and compare results on their own tasks.

### A monthly allowance for experiments

[15:07](https://www.youtube.com/watch?v=btxGmN8RvNU&t=907s)

The closing question resolves the unit behind the free offer: is the allowance daily? Levi states that the free tier includes 5,000 requests per month. He also mentions pay-as-you-go without giving prices or detailed conditions. His adoption context is an MVP, a small experiment, or a hackathon: connect the agent to MCP and use that access while building the application. The recording ends with that invitation, rather than an installation walkthrough or a production deployment.

## Resources

- [Bright Data MCP](https://github.com/brightdata/brightdata-mcp) — Server setup, retrieval tools, browser automation, tool filtering, and current free-tier documentation.
- [Bright Data Agent Skills](https://github.com/brightdata/skills) — Agent skills for building scrapers, collecting structured web data, and orchestrating MCP tools.
- [Cloudflare's AI Labyrinth announcement](https://blog.cloudflare.com/ai-labyrinth/) — Explains the opt-in crawler honeypot and its use of irrelevant generated pages.
- [AI Search Has a Citation Problem](https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php) — A study of eight AI search tools identifying news articles from excerpts, with methodology and citation-error results.
- [Privacy authorities on data scraping](https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2024/10/global-privacy-authorities-issue-follow-up-joint-statement-on-data-scraping-after-industry-engagement/) — Regulatory guidance on lawful collection and protection of publicly accessible personal information.

## Slides

### AI Engineer Europe

[0:00](https://www.youtube.com/watch?v=btxGmN8RvNU&t=0s) [AI Engineer Europe logo on a black background.](https://assets.ai.engineer/sites/aiecode2025/sha256/24e440bf5dfd4ecccd4fffdc9ef9243be07fff10b16e87c6b80fbc594828e3ea.webp)

AI Engineer
EUROPE

### Presenting Sponsor

Google DeepMind is the presenting sponsor.

[0:05](https://www.youtube.com/watch?v=btxGmN8RvNU&t=5s) [Google DeepMind presenting sponsor card.](https://assets.ai.engineer/sites/aiecode2025/sha256/e0a2733394b9e76d5cbcc292edcb1e7d0d4305725368334cfa74aa4dc6f38f30.webp)

PRESENTING SPONSOR

Google DeepMind

### Platinum Sponsors

Braintrust, WorkOS and OpenAI are the platinum sponsors.

[0:10](https://www.youtube.com/watch?v=btxGmN8RvNU&t=10s) [Platinum sponsors Braintrust, WorkOS and OpenAI.](https://assets.ai.engineer/sites/aiecode2025/sha256/8f633e39687b44d51a588ef4b9521d4ac17851e980fbbafc17cbed015e632d87.webp)

PLATINUM SPONSORS

Braintrust
WorkOS
OpenAI

### Your Agent's Biggest Lie: I Searched the Web

Rafael Levi introduces his concern that LLMs can claim to have searched the web when they have not.

[0:26](https://www.youtube.com/watch?v=btxGmN8RvNU&t=26s) [Bright Data title slide introducing Rafael Levi's talk about agents searching the web, with a promotional QR code and an illustration.](https://assets.ai.engineer/sites/aiecode2025/sha256/ed26fa2f7cdd83a338649a6739d4a671be9e2f334eeb84ec10720ac8a734b40d.webp)

bright data | AI Engineer EUROPE

SESSION 1

Your Agent's Biggest Lie:
“I SEARCHED THE WEB”

Rafael Levi
DevRel Engineer | Bright Data

Sign up and get a bonus

Interface text:
Session1_Agents_Biggest_Lie
PPTX
Working offline
File  Edit  View  Insert  Format  Slide  Arrange  Tools  Extensions  Help
Slideshow
Start slideshow (Ctrl+F5)
Fit  Background  Layout  Theme  Transition
Enhance this slide
TIMING: 30 seconds
Search
19°C Partly sunny
11:00 AM
4/9/2026
AIE

### AI agents get blocked

Rafael Levi says agents may encounter CAPTCHAs without reporting them, then present training data as current information. The slide also warns of blocking and fake content.

[0:54](https://www.youtube.com/watch?v=btxGmN8RvNU&t=54s) [Slide states that AI agents encounter blocking, fake content and CAPTCHAs, then may report as if nothing went wrong.](https://assets.ai.engineer/sites/aiecode2025/sha256/6ca74bddbd746cc2a665074351a530dd4e89f8562bba26d1796d1069869412d2.webp)

AI agents get blocked,
fed fake content, and hit CAPTCHAs.
Then they report back as if nothing went wrong.

bright data
2

Interface text:
Session1_Agents_Biggest_Lie
PPTX
Working offline
File Edit View Insert Format Slide Arrange Tools Extensions Help
Fit
Background Layout Theme Transition
Slideshow
Share
Enhance this slide
Click to add speaker notes
AIE
Braintrust
WorkOS
OpenAI

[1:08](https://www.youtube.com/watch?v=btxGmN8RvNU&t=68s) [Slide says AI agents encounter blocks, fake content and CAPTCHAs, then report back as if nothing went wrong.](https://assets.ai.engineer/sites/aiecode2025/sha256/923357e4c15406ce6efa89df44bf3a2388490d1739e170db4f071df5730039cb.webp)

AI agents get blocked,
fed fake content, and hit CAPTCHAs.
Then they report back as if nothing went wrong.

bright data
2

Interface text:
Session1_Agents_Biggest_Lie
PPTX
Working offline
File Edit View Insert Format Slide Arrange Tools Extensions Help
Fit
Background Layout Theme Transition
Slideshow
Share
Enhance this slide
Click to add speaker notes

### The Web Is Fighting Your Agent

The slide says Cloudflare blocks AI crawlers on 20% of the web, while its AI Labyrinth feeds agents plausible-looking fake content.

[1:38](https://www.youtube.com/watch?v=btxGmN8RvNU&t=98s) [Slide contrasts AI crawlers being blocked by default with agents receiving fake content.](https://assets.ai.engineer/sites/aiecode2025/sha256/b5798473aaf15c9feabd69eadbef770b747efc9db4facb9078c0b39453b871a2.webp)

The Web Is Fighting Your Agent
And your agent doesn't tell you it lost.

Blocked by Default
Cloudflare blocks AI crawlers on 20% of the web. No config needed by site owners. Your agent hits a wall before it starts.

20% of the web blocks AI crawlers

Actively Poisoned
Cloudflare's AI Labyrinth feeds fake content to AI agents. Your agent gets data back — it's just not real. Plausible-looking disinformation.

FAKE content served to AI agents

bright data
3

Interface text:
docs.google.com/presentation/d/[illegible]
Ask Google
Claude
Claude (MCP)
Cy
Flowers
All Bookmarks
Session1_Agents_Biggest_Lie
PPTX
File Edit View Insert Format Slide Arrange Tools Extensions Help
Fit
Background Layout Theme Transition
Slideshow
Share
Enhance this slide
TIMING: 90 seconds
Search
19°C
Partly sunny
ENG
11:01 AM
4/9/2026

### The Invisible Failure Loop

An agent sends a request, encounters an anti-bot challenge or unusable content, and may make up an answer while reporting success without warning.

[2:05](https://www.youtube.com/watch?v=btxGmN8RvNU&t=125s) [Four-step diagram shows an agent request intercepted by anti-bot defenses, receiving unusable content and reporting success.](https://assets.ai.engineer/sites/aiecode2025/sha256/da0ed322dbddf5e60575c58c7bf859b7c5794900008b8ebc8ca43284524052ed.webp)

The Invisible Failure Loop
No errors. No warnings. Just wrong answers presented as truth.

1. Agent Sends Request
Looks like a normal HTTP request

2. Anti-Bot Intercepts
CAPTCHA, JS challenge, or AI Labyrinth

3. Agent Gets Garbage Back
Empty page, fake content, or challenge HTML

4. Agent Reports Success
Hallucinates answer from whatever it received

bright data
4

Interface text:
docs.google.com/presentation/d/[illegible]/edit?slide=id.p4#slide=id.p4
Session1_Agents_Biggest_Lie
PPTX
File Edit View Insert Format Slide Arrange Tools Extensions Help
Fit
Background Layout Theme Transition
Slideshow Share
Enhance this slide
TIMING: 60 seconds
19°C Partly sunny
11:01 AM
4/9/2026

### The Evidence

The slide attributes a 60%+ citation failure rate to a Columbia/Tow Center test of eight AI search engines: cited content was never retrieved. The speaker describes product links that lead nowhere.

[2:36](https://www.youtube.com/watch?v=btxGmN8RvNU&t=156s) [Slide claims a citation failure rate above 60% in a study of eight AI search engines.](https://assets.ai.engineer/sites/aiecode2025/sha256/938489a14901e1f21e39a782b9696a8fd276f12b7b08e7676a23d8d4c27f3068.webp)

The Evidence

60%+
citation failure rate

Columbia / Tow Center for Digital Journalism
tested 8 major AI search engines.

More than 60% of citations referenced content the AI never successfully retrieved.

This isn't a fringe edge case. This is the majority of AI-generated answers citing content that was never accessed.

bright data
5

AIE
Engineering the future of AI

Interface text:
Session1_Agents_Biggest_Lie
PPTX
docs.google.com/presentation/d/[illegible]/edit?slide=id.p3#slide=id.p3
Ask Google
File  Edit  View  Insert  Format  Slide  Arrange  Tools  Extensions  Help
Slideshow  Share
Fit  Background  Layout  Theme  Transition
Enhance this slide
TIMING: 60 seconds
19°C  Partly sunny
11:02 AM  4/9/2026

### Live Demo Comparison

The presenter plans to compare the same agent on the same sites with and without Bright Data Web MCP. The slide predicts blocked requests and fabricated data in one case, and verified data in the other.

[3:00](https://www.youtube.com/watch?v=btxGmN8RvNU&t=180s) [Slide compares claimed agent outcomes without web infrastructure and with Bright Data Web MCP.](https://assets.ai.engineer/sites/aiecode2025/sha256/42f1ca942a14f2007454da3468058c3aefaf708bb1216533814a24d5620b3939.webp)

Live Demo
Same agent. Same sites. Two very different outcomes.

Without Web Infrastructure
• Raw agent HTTP requests
• Hits Cloudflare walls
• JavaScript doesn't render
• CAPTCHAs block silently
• Returns fabricated data

With Bright Data Web MCP
• Same agent, same code
• Anti-bot bypass handled
• Full JS rendering
• CAPTCHAs solved automatically
• Returns real, verified data

bright data
6

Interface text:
Session1_Agents_Biggest_Lie
PPTX
File Edit View Insert Format Slide Arrange Tools Extensions Help
Slideshow Share
Fit Background Layout Theme Transition
Enhance this slide
TIMING: 30 seconds
AIE
Engineering the future of AI

### Demo task setup

The demo gives the same five site-specific tasks to scripts with and without Bright Data MCP, from rental listings to a TikTok profile.

[3:21](https://www.youtube.com/watch?v=btxGmN8RvNU&t=201s) [VS Code shows five web-data tasks for the demo.](https://assets.ai.engineer/sites/aiecode2025/sha256/e42d4456585cbdd6e061d91d05458cd4878e11b3548a779a96f58548d86ed4ea.webp)

# IDENTICAL task + system prompt to Demo 2

TASK = """
I need you to complete these 5 tasks. For each one, try your best to get the data. If you can't, explain what went wrong.

1. PROPERTY — Get rental listings near the conference venue in London from Rightmove:
   https://www.rightmove.co.uk/property-to-rent/find.html?searchLocation=SW1%2C+South+West+London&useLocationIdentifier=1[illegible]
   List at least 5 specific properties with prices.

2. COMPANY — Get the LinkedIn company profile for Mistral AI:
   https://www.linkedin.com/company/mistral-ai/

3. PROFILES — Get the Instagram profile for @ylecun:
   https://www.instagram.com/ylecun/

4. PRODUCT — Get the Amazon product page for this book:
   https://www.amazon.com/dp/1098107969

5. SOCIAL — Get the TikTok profile for @google:
   https://www.tiktok.com/@google

Interface text:
WORKSHOP
.env
.env.example
01_without_mcp.py
02_with_mcp.py
03_live_challenge.py
README.md
requirements.txt
results_comparison.md
results_without_mcp.md
PROBLEMS OUTPUT DEBUG CONSOLE TERMINAL PORTS
Update is ready, click to restart.
AIE
AI Engineer EUROPE

### Demo without web tools: scorecard

GPT-5, run without browsing tools, reports that all five web-data tasks failed. The demo then directs the presenter to run the same tasks with Bright Data MCP tools.

[4:21](https://www.youtube.com/watch?v=btxGmN8RvNU&t=261s) [The demo output reports zero successes and five failures without web tools.](https://assets.ai.engineer/sites/aiecode2025/sha256/91e8ffcba690e638f85f7e684a5b248299416c7a79a2b71a85197009ef91d4f8.webp)

[illegible]

2. COMPANY — Get the LinkedIn company profile for Mistral AI:
    https://www.linkedin.com/company/mistral-ai/

3. PROFILES — Get the Instagram profile for @ylecun:
    https://www.instagram.com/ylecun/

4. PRODUCT — Get the Amazon product page for this book:
    https://www.amazon.com/dp/1098107969

5. SOCIAL — Get the TikTok profile for @google:
    https://www.tiktok.com/@google

After all 5 tasks, give a final scorecard: how many succeeded vs failed,
keep in mind if you found data on a different site it does not count as success.
"""

SYSTEM_PROMPT = (

- Failed: Amazon pages are protected by bot checks and dynamic rendering; without live web access I can’t retrieve https://www.amazon.com/dp/1098107969. If you share the HTML or enable browsing, I can parse title, price, rating, and details.

5) SOCIAL — TikTok profile for @google
- Failed: TikTok heavily relies on JavaScript and region/CDN gating; without a browser tool I can’t load https://www.tiktok.com/@google. If you provide a page export, I can extract bio, follower counts, and recent videos.

Scorecard: 0 succeeded, 5 failed.

Results saved to: c:\Users\skurg\Desktop\Coding\Git-Hub\brightdata-hack-pack\workshop\results_without_mcp.md

Now run: python 02_with_mcp.py
Same 5 tasks — but with Bright Data MCP tools.

[Done] exited with code=0 in 28.707 seconds

AIE
Engineering the future of AI

Interface text:
File Edit Selection View Go Run Terminal Help
workshop
EXPLORER
WORKSHOP
.env
.env.example
01_without_mcp.py
02_with_mcp.py
03_live_challenge.py
README.md
requirements.txt
results_comparison.md
results_without_mcp.md
PROBLEMS OUTPUT DEBUG CONSOLE TERMINAL PORTS
Code
OUTLINE
TIMELINE
Update is ready, click to restart
Ln 40, Col 20
Spaces: 4
UTF-8
CRLF
Python
Python 3.13.4
Continue (NE)
19°C Partly sunny
11:03 AM
4/9/2026

### Demo without web tools: agent response

With no tools available, GPT-5 says it cannot retrieve Rightmove rentals or Mistral AI’s LinkedIn profile. The presenter reports that all five tasks failed.

[4:32](https://www.youtube.com/watch?v=btxGmN8RvNU&t=272s) [The agent explains why it could not retrieve the rental listings and company profile.](https://assets.ai.engineer/sites/aiecode2025/sha256/53d004fb076ec960d283580b22bf7e176850525714e0579fe107c37866a2e6e8.webp)

2. COMPANY — Get the LinkedIn company profile for Mistral AI:
https://www.linkedin.com/company/mistral-ai/

3. PROFILES — Get the Instagram profile for @ylecun:
https://www.instagram.com/ylecun/

4. PRODUCT — Get the Amazon product page for this book:
https://www.amazon.com/dp/1098107969

5. SOCIAL — Get the TikTok profile for @google:
https://www.tiktok.com/@google

After all 5 tasks, give a final scorecard: how many succeeded vs failed,
keep in mind if you found data on a different site it does not count as success.
"""

SYSTEM_PROMPT = (

Tasks:
[illegible]
4. Product (Amazon)
5. Social (TikTok)

Tools available: NONE
Calling GPT-5...

AGENT RESPONSE (24.7s)

1) PROPERTY — Rightmove (London SW1, rentals)
- Failed: I don’t have live web access in this chat, and Rightmove’s listings are rendered behind JavaScript/CDN protections (Akamai) that require a real browser session. Without a browsing tool or page HTML, I can’t extract at least 5 specific properties with prices.

2) COMPANY — LinkedIn company profile for Mistral AI
- Failed: LinkedIn company pages are login-gated and protected by anti-bot measures; without an authenticated browser tool I can’t load or parse https://www.linkedin.com/company/mistral-ai/. If you can enable browsing or share the page HTML, I’ll extract the [illegible]

AIE
Engineering the future of AI

Interface text:
EXPLORER
WORKSHOP
.env
.env.example
01_without_mcp.py
02_with_mcp.py
03_live_challenge.py
README.md
requirements.txt
results_comparison.md
results_without_mcp.md
PROBLEMS  OUTPUT  DEBUG CONSOLE  TERMINAL  PORTS
OUTLINE  TIMELINE
Update is ready, click to restart

### MCP connection and tools

The presenter reruns the five tasks with Bright Data MCP. The connection reports 69 tools, including search and Markdown scraping; the task results are not yet shown.

[4:56](https://www.youtube.com/watch?v=btxGmN8RvNU&t=296s) [Demo output lists five tasks and reports a connection with 69 available tools.](https://assets.ai.engineer/sites/aiecode2025/sha256/ab82177069d882647cf8f22b0f5abc36196484d9b8fe87e05871da4e0ecf0aab.webp)

https://www.rightmove.co.uk/property-to-rent/find.html?searchLocation=SW1%2C+South+West+London&useLocationIdentifier=[illegible]
List at least 5 specific properties with prices.

2. COMPANY – Get the LinkedIn company profile for Mistral AI:
https://www.linkedin.com/company/mistral-ai/

3. PROFILES – Get the Instagram profile for @ylecun:
https://www.instagram.com/ylecun/

Tasks:
1. Property (Rightmove)
2. Company (LinkedIn Company)
3. Profiles (Instagram)
4. Product (Amazon)
5. Social (TikTok)

Connecting to Bright Data MCP server via SSE...
Connected! 69 tools available:
- search_engine
- scrape_as_markdown
- search_engine_batch
- scrape_batch
- scrape_as_html
- discover
- session_stats
- web_data_amazon_product
- web_data_amazon_product_reviews
- web_data_amazon_product_search
- web_data_walmart_product
- web_data_walmart_seller
- web_data_ebay_product
- web_data_homedepot_products
- web_data_zara_products
- web_data_etsy_products
- web_data_bestbuy_products
- web_data_linkedin_person_profile

AIE
Engineering the future of AI

Interface text:
File Edit Selection View Go Run Terminal Help
WORKSHOP
.env
.env.example
01_without_mcp.py
02_with_mcp.py
03_live_challenge.py
README.md
requirements.txt
results_comparison.md
results_without_mcp.md
PROBLEMS OUTPUT DEBUG CONSOLE TERMINAL PORTS
OUTLINE TIMELINE
Update is ready, click to restart.
Ln 89, Col 38  Spaces: 4  UTF-8  CRLF  Python  Python 3.13.4
19°C  Partly sunny  11:04 AM  4/9/2026

### Bright Data MCP five-task demonstration

The demo lists 69 available MCP tools, then shows successful retrievals from Rightmove, Mistral AI’s LinkedIn profile, Instagram and Amazon. The output reports response sizes and elapsed times for each call.

[5:44](https://www.youtube.com/watch?v=btxGmN8RvNU&t=344s) [Demo output lists tools for company profiles, social platforms, shopping, and other web data.](https://assets.ai.engineer/sites/aiecode2025/sha256/1166793689f0340e2b92b781d9a84068f7c4cd53ecc978d84779d22bae66c6ed.webp)

https://www.rightmove.co.uk/property-to-rent/find.html?searchLocation=SW1%2C+South+West+London&useLocationIdentifier=[illegible]
List at least 5 specific properties with prices.

2. COMPANY — Get the LinkedIn company profile for Mistral AI:
https://www.linkedin.com/company/mistral-ai/

3. PROFILES — Get the Instagram profile for @ylecun:
https://www.instagram.com/ylecun/

Connected! 69 tools available:
- web_data_linkedin_person_profile
- web_data_linkedin_company_profile
- web_data_linkedin_job_listings
- web_data_linkedin_posts
- web_data_linkedin_people_search
- web_data_crunchbase_company
- web_data_zoominfo_company_profile
- web_data_instagram_profiles
- web_data_instagram_posts
- web_data_instagram_reels
- web_data_instagram_comments
- web_data_facebook_posts
- web_data_facebook_marketplace_listings
- web_data_facebook_company_reviews
- web_data_facebook_events
- web_data_tiktok_profiles
- web_data_tiktok_posts
- web_data_tiktok_shop
- web_data_tiktok_comments
- web_data_google_maps_reviews
- web_data_google_shopping
- web_data_google_play_store
- web_data_apple_app_store
- web_data_reuter_news
- web_data_github_repository_file
- web_data_yahoo_finance_business
- web_data_x_posts
- web_data_zillow_properties_listing

Interface text:
WORKSHOP
.env
.env.example
01_without_mcp.py
02_with_mcp.py
03_live_challenge.py
README.md
requirements.txt
results_comparison.md
results_without_mcp.md
PROBLEMS  OUTPUT  DEBUG CONSOLE  TERMINAL  PORTS
AIE
AI Engineer EUROPE

[6:15](https://www.youtube.com/watch?v=btxGmN8RvNU&t=375s) [A code editor shows successful property, company and social profile data retrieval.](https://assets.ai.engineer/sites/aiecode2025/sha256/f384ae1575f9357a04ae706d8ce74bda6540d92e9d56f87aca748f831d21b052.webp)

https://www.rightmove.co.uk/property-to-rent/find.html?searchLocation=SW1%2C+South+West+London&useLocationIdentifier=[illegible]
List at least 5 specific properties with prices.

2. COMPANY — Get the LinkedIn company profile for Mistral AI:
https://www.linkedin.com/company/mistral-ai/

3. PROFILES — Get the Instagram profile for @ylecun:
https://www.instagram.com/ylecun/

BRIGHT DATA MCP: scrape_as_markdown
Args: {"url": "https://www.rightmove.co.uk/property-to-rent/find.html?searchLocation=SW1%2C+South+West+..."}
Result: SUCCESS | 41,657 chars | 2.4s
Preview: Properties To Rent in SW1 | Rightmoveopen-rightmove-assistant burgermenuBuyProperty for saleNew homes for saleProperty valuationAgent valuationInvestorsMortgagesRentProperty to rentStudent property to...

BRIGHT DATA MCP: web_data_linkedin_company_profile
Args: {"url": "https://www.linkedin.com/company/mistral-ai/"}
Result: SUCCESS | 33,371 chars | 5.6s
Preview: [{"id":"mistralai","name":"Mistral AI","countrycode":"FR","locations":["Paris, France 75010, FR"],"followers":622891,"employeesinlinkedin":970,"about":"Frontier AI. In your hands. We believe in a futu...

BRIGHT DATA MCP: web_data_instagram_profiles
Args: {"url": "https://www.instagram.com/ylecun/"}
Result: SUCCESS | 3,268 chars | 2.6s
Preview: [{"account":"ylecun","fbid":"17841401555783297","id":"851059859","followers":2716,"postscount":3,"isbusinessaccount":false,"isprofessionalaccount":false,"isverified":false,"avgengagement":0.0384,"exte...

BRIGHT DATA MCP: web_data_amazon_product
Args: {"url": "https://www.amazon.com/dp/1098107969"}
Result: SUCCESS | 12,971 chars | 8.5s

Interface text:
WORKSHOP
.env
.env.example
01_without_mcp.py
02_with_mcp.py
03_live_challenge.py
README.md
requirements.txt
results_comparison.md
results_without_mcp.md
PROBLEMS  OUTPUT  DEBUG CONSOLE  TERMINAL  PORTS
AIE
AI Engineer EUROPE

### Public company profile

The demo opens Mistral AI’s public company profile. The presenter says profile access can eventually require a login and that the service collects public data.

[8:06](https://www.youtube.com/watch?v=btxGmN8RvNU&t=486s) [Mistral AI's public LinkedIn company profile, including its overview and company information.](https://assets.ai.engineer/sites/aiecode2025/sha256/e4fa460bb32efb26f75d8a6f28fb0bcf446947cba476c40e0b5bfdbe1ae08833.webp)

Mistral AI
Technology, Information and Internet
Paris, France · 622,893 followers
Frontier AI. In Your Hands.
Discover all 970 employees
See jobs
Follow

About us
Frontier AI. In your hands. We believe in a future where AI is abundant and accessible. We aspire to empower the world to build with—and benefit from—the most significant technology of our time. Join us: https://jobs.lever.co/mistral/

Website
https://mistral.ai

Similar pages
Anthropic
Research Services
Hugging Face
Software Development
Perplexity
Software Development
San Francisco, California
ElevenLabs
Research Services
Show more similar pages

Browse jobs
Scientist jobs
6,219 open jobs
Analyst jobs
3,411 open jobs
Marketing Manager jobs
3,742 open jobs
Researcher jobs

Interface text:
fr.linkedin.com/company/mistralai
Incognito
LinkedIn respects your privacy
LinkedIn and 3rd parties use essential and non-essential cookies to provide, secure, analyze and improve our Services, and to show you relevant ads (including professional and job ads) on and off LinkedIn. Learn more in our Cookie Policy.
Select Accept to consent or Reject to decline non-essential cookies for this use. You can update your choices at any time in your settings.
Accept
Reject
LinkedIn
Top Content
People
Learning
Jobs
Games
Get the app
Sign in
Join now
LinkedIn is better on the app
Don’t have the app? Get it in the Microsoft Store.
Open the app
Search
19°C
Partly sunny
ENG
11:07 AM
4/9/2026

### Live Demo

The slide contrasts raw agent requests that encounter access barriers with the same agent using Bright Data Web MCP, which it says handles anti-bot checks and returns verified data.

[8:21](https://www.youtube.com/watch?v=btxGmN8RvNU&t=501s) [Slide compares agent web access without infrastructure and with Bright Data Web MCP.](https://assets.ai.engineer/sites/aiecode2025/sha256/64984d456070bfd69643dca7fe0958bfd50603827b5dcc3dc94391a17a789b52.webp)

Live Demo
Same agent. Same sites. Two very different outcomes.

Without Web Infrastructure
• Raw agent HTTP requests
• Hits Cloudflare walls
• JavaScript doesn't render
• CAPTCHAs block silently
• Returns fabricated data

With Bright Data Web MCP
• Same agent, same code
• Anti-bot bypass handled
• Full JS rendering
• CAPTCHAs solved automatically
• Returns real, verified data

bright data
6

Interface text:
Session1_Agents_Biggest_Lie
File  Edit  View  Insert  Format  Slide  Arrange  Tools  Extensions  Help
Background  Layout  Theme  Transition
Slideshow  Share
Enhance this slide
02_with_mcp.py - works...
TIMING: 30 seconds
19°C  Partly sunny
11:07 AM  4/9/2026

### Head-to-Head Comparison

The LLM’s scorecard reports 0/5 successful web tasks without MCP and 5/5 with MCP, across property listings, company and social profiles, and a product page. These are the demo’s reported results, not independent verification.

[8:29](https://www.youtube.com/watch?v=btxGmN8RvNU&t=509s) [Code editor output compares five web-access tasks and reports zero successes without MCP and five with MCP.](https://assets.ai.engineer/sites/aiecode2025/sha256/790f5c25687d48d65dff36106b0686862defda81bb6c4af08e3b686b4761d9c3.webp)

https://www.rightmove.co.uk/property-to-rent/find.html?searchLocation=SW1%2C+South+West+London&useLocationIdentifier=[illegible]
List at least 5 specific properties with prices.

2. COMPANY — Get the LinkedIn company profile for Mistral AI:
https://www.linkedin.com/company/mistral-ai/

HEAD-TO-HEAD COMPARISON

Comparison scorecard

1) Property — Rightmove (London SW1 rentals)
- Demo 1: Failed — no live web access; JS/CDN (Akamai) blocked; 0 listings.
- Demo 2: Succeeded — 5+ listings with addresses and prices (e.g., Finlay St £12,000 pcm; Coppetts Rd £10,000 pcm; Goldhawk Rd £6,700 pcm; Birkbeck Rd £3,200 pcm; Campen Close £2,750 pcm).

2) Company — LinkedIn (Mistral AI)
- Demo 1: Failed — login-gated + anti-bot; no profile data.
- Demo 2: Succeeded — URL + 622,891 followers; size 201-500; founded 2023; industry; HQ Paris; website; tagline.

3) Profiles — Instagram (@ylecun)
- Demo 1: Failed — login required + heavy client-side rendering.
- Demo 2: Succeeded — Name, followers 2,716, posts 3, bio, external link; public, not verified.

4) Product — Amazon (ASIN 1098107969)
- Demo 1: Failed — bot checks/dynamic HTML; no product details.
- Demo 2: Succeeded — Title, rating 4.6/5 (939), price $40 paperback, formats (Kindle, Audiobook, Spiral-bound).

5) Social — TikTok (@google)
- Demo 1: Failed — JS/region gating; no stats.
- Demo 2: Succeeded — Verified; 3.3M followers; 16.8M likes; 834 videos; bio + link; recent post themes.

Final verdict
- Demo 1: 0/5 succeeded
- Demo 2: 5/5 succeeded

Interface text:
WORKSHOP
.env
.env.example
01_without_mcp.py
02_with_mcp.py
03_live_challenge.py
README.md
requirements.txt
results_comparison.md
results_without_mcp.md
PROBLEMS  OUTPUT  DEBUG CONSOLE  TERMINAL  PORTS
AI Engineer EUROPE

### Recap

The presenter warns that blocked agents can make up answers and that AI Labyrinth can feed them fake data. He presents Web MCP as a way to handle access barriers.

[9:49](https://www.youtube.com/watch?v=btxGmN8RvNU&t=589s) [Recap slide lists silent agent failures, fake content, and web infrastructure as three takeaways.](https://assets.ai.engineer/sites/aiecode2025/sha256/ab204eef045981703a6f393ad73728eed77a72d9badaa9b6f3d5b54d6a32c60c.webp)

Recap

01 Your Agent is failing silently
60%+ citation failure is not an exception.. No errors thrown. No retries triggered. The agent genuinely believes it has the data.

02 Fake Content Is an Active Threat
Cloudflare AI Labyrinth feeds your agent disinformation. The data LOOKS real. It’s not. This is adversarial by design.

03 The Fix Is Infrastructure, Not Code
You can’t code around anti-bot systems. Web MCP handles IP rotation, CAPTCHA solving, and JS rendering — transparently, with minimal code change.

bright data
8

AIE
Engineering the future of AI

Interface text:
Session1_Agents_Biggest_Lie
File Edit View Insert Format Slide Arrange Tools Extensions Help
Background Layout Theme Transition
Slideshow Share
LIVE DEMO
Stop Trusting. Start Verifying.
Thank you!
Enhance this slide
TIMING: 60 seconds

[10:19](https://www.youtube.com/watch?v=btxGmN8RvNU&t=619s) [Recap slide asserting that agents can fail silently, encounter fake content, and benefit from web access infrastructure.](https://assets.ai.engineer/sites/aiecode2025/sha256/6291ad7e23a59e3deb9e3f90ae1860f7419953ccf69b414f09e103c074400657.webp)

Recap

01 Your Agent is failing silently
60%+ citation failure is not an exception.. No errors thrown. No retries triggered. The agent genuinely believes it has the data.

02 Fake Content Is an Active Threat
Cloudflare AI Labyrinth feeds your agent disinformation. The data LOOKS real. It’s not. This is adversarial by design.

03 The Fix Is Infrastructure, Not Code
You can’t code around anti-bot systems. Web MCP handles IP rotation, CAPTCHA solving, and JS rendering — transparently, with minimal code change.

bright data
8

Interface text:
Session1_Agents_Biggest_Lie
File Edit View Insert Format Slide Arrange Tools Extensions Help
Background Layout Theme Transition
Slideshow Share
LIVE DEMO
Stop Trusting. Start Verifying.
Thank you!
Enhance this slide
TIMING: 60 seconds

### Stop Trusting. Start Verifying.

Bright Data invites attendees to try Web MCP. The speaker says it has a free tier of 5,000 requests, but initially is unsure where the slide’s QR code leads.

[10:26](https://www.youtube.com/watch?v=btxGmN8RvNU&t=626s) [Bright Data Web MCP call-to-action slide with a QR code and a note about the next session.](https://assets.ai.engineer/sites/aiecode2025/sha256/62d0cd5759afbdeeec68b43b31ac72c207af0b2e24eb159e1339706f55b99a47.webp)

Stop Trusting.
Start Verifying.

Try Bright Data Web MCP

Works with any MCP-compatible agent.
Integrate in minutes. Verify every request.

Next session:
From MCP to Scale: when you need to go from one URL to a full production pipeline.

bright data
9

Interface text:
Session1_Agents_Biggest_Lie
PPTX
File Edit View Insert Format Slide Arrange Tools Extensions Help
Slideshow
Share
Fit
Background Layout Theme Transition
Enhance this slide
TIMING: 30 seconds
Search
19°C Partly sunny
11:09 AM
4/9/2026

### Bright Data skills repository

The presenter points to Bright Data’s skills repository, which he says teaches agents to build scrapers and data-collection pipelines. He plans to demonstrate a pipeline in a later session.

[12:45](https://www.youtube.com/watch?v=btxGmN8RvNU&t=765s) [GitHub repository page showing the Bright Data Plugin for Claude Code.](https://assets.ai.engineer/sites/aiecode2025/sha256/95f74884865583641b657801b2b8846e51e6de7e9a5d459e8566b65b03ba2f0d.webp)

brightdata / skills
Public

.claude-plugin — feat: add competitive intel
skills — fix:remove redundant files
.gitignore — fix: enhance mcp skill
LICENSE — add bd plugin
README.md — feat: add competitive intel

README
Bright Data Plugin for Claude Code
Unlock the web with AI-powered scraping, search, and structured data

POWERED BY BRIGHT DATA
LICENSE MIT
SKILLS 11
DATASETS 40+
MCP TOOLS 60+

Quick Start · Skills · Data Feeds · MCP · CLI · Competitive Intel · Scraper Builder · Best Practices · Python SDK · Setup

About
skills.sh/brightdata
Readme
MIT license
Activity
Custom properties
20 stars
0 watching
5 forks
Report repository

Releases
No releases published

Packages
No packages published

Contributors 3
meirk-brd Meir Kadosh
claude Claude
EyalDassa

AI Engineer EUROPE
AIE

Interface text:
github.com/brightdata/skills
Platform · Solutions · Resources · Open Source · Enterprise · Pricing
Search or jump to...
Sign in · Sign up
Notifications · Fork 5 · Star 20
Code · Issues 1 · Pull requests · Actions · Projects · Security and quality · Insights
main · 2 Branches · 0 Tags · Go to file · Code
meirk-brd · fix:remove redundant files · last week · 24 Commits
README · MIT license
19°C Partly sunny
11:12 AM · 4/9/2026

### Mistral AI company page

The Mistral AI LinkedIn profile is open as an audience member asks whether the earlier experiment used OpenAI web search or a call to Bright Data.

[14:11](https://www.youtube.com/watch?v=btxGmN8RvNU&t=851s) [LinkedIn company page for Mistral AI with a cookie notice above the profile.](https://assets.ai.engineer/sites/aiecode2025/sha256/4b162f47d9c822bdee8e782a808188c5d60cb66a4acdb5065cc1c27ee7ddcefc.webp)

Mistral AI
Technology, Information and Internet
Paris, France · 622,893 followers
Frontier AI. In Your Hands.
Discover all 970 employees
See jobs
Follow

About us
Frontier AI. In your hands. We believe in a future where AI is abundant and accessible. We aspire to empower the world to build with—and benefit from—the most significant technology of our time. Join us: https://jobs.lever.co/mistral/
Website: https://mistral.ai

Similar pages
Anthropic — Research Services
Hugging Face — Software Development
Perplexity — Software Development — San Francisco, California
ElevenLabs — Research Services
Show more similar pages

Browse jobs
Scientist jobs — 6,219 open jobs
Analyst jobs — 3,411 open jobs
Marketing Manager jobs — 3,742 open jobs
Researcher jobs

Interface text:
fr.linkedin.com/company/mistralai
LinkedIn respects your privacy
LinkedIn and 3rd parties use essential and non-essential cookies to provide, secure, analyze and improve our Services, and to show you relevant ads (including professional and job ads) on and off LinkedIn. Learn more in our Cookie Policy.
Select Accept to consent or Reject to decline non-essential cookies for this use. You can update your choices at any time in your settings.
Accept
Reject
LinkedIn
Top Content
People
Learning
Jobs
Games
Get the app
Sign in
Join now
LinkedIn is better on the app
Don’t have the app? Get it in the Microsoft Store.
Open the app
AIE
Google DeepMind

### Web data demo tasks and results

The presenter asked the demos to load five specific URLs, without using search. The displayed scorecard reports 0/5 successes for Demo 1 and 5/5 for Demo 2.

[14:35](https://www.youtube.com/watch?v=btxGmN8RvNU&t=875s) [Code editor showing five web data tasks and output reporting zero of five successes for Demo 1 and five of five for Demo 2.](https://assets.ai.engineer/sites/aiecode2025/sha256/173d70dc48dba59127c405b485eaa32159c8a74ca8ab68fd6fdebab621db288e.webp)

the data. If you can't, explain what went wrong.

1. PROPERTY — Get rental listings near the conference venue in London from Rightmove:
https://www.rightmove.co.uk/property-to-rent/find.html?searchLocation=SW1%2C+South+West+London&useLocationIdentifier=[illegible]
List at least 5 specific properties with prices.

2. COMPANY — Get the LinkedIn company profile for Mistral AI:
https://www.linkedin.com/company/mistral-ai/

3. PROFILES — Get the Instagram profile for @ylecun:
https://www.instagram.com/ylecun/

4. PRODUCT — Get the Amazon product page for this book:
https://www.amazon.com/dp/1098107969

5. SOCIAL — Get the TikTok profile for @google:
https://www.tiktok.com/@google

After all 5 tasks, give a final scorecard: how many succeeded vs failed,
keep in mind if you found data on a different site it does not count as success.
"""

SYSTEM_PROMPT = [illegible]

Output:
- Demo 1: Failed — login required + heavy client-side rendering.
- Demo 2: Succeeded — Name, followers 2,716, posts 3, bio, external link; public, not verified.

4) Product — Amazon (ASIN 1098107969)
- Demo 1: Failed — bot checks/dynamic HTML; no product details.
- Demo 2: Succeeded — Title, rating 4.6/5 (939), price $40 paperback, formats (Kindle, Audiobook, Spiral-bound).

5) Social — TikTok (@google)
- Demo 1: Failed — JS/region gating; no stats.
- Demo 2: Succeeded — Verified; 3.3M followers; 16.8M likes; 834 videos; bio + link; recent post themes.

Final verdict
- Demo 1: 0/5 succeeded
- Demo 2: 5/5 succeeded

Interface text:
File Edit Selection View Go Run Terminal Help
WORKSHOP
.env
.env.example
01_without_mcp.py
02_with_mcp.py
03_live_challenge.py
README.md
requirements.txt
results_comparison.md
results_without_mcp.md
PROBLEMS OUTPUT DEBUG CONSOLE TERMINAL PORTS
OUTLINE TIMELINE
Update is ready, click to restart.
Ln 40, Col 20  Spaces: 4  UTF-8  CRLF  Python  Python 3.13.4  Continue (NE)
19°C Partly sunny  11:14 AM  4/9/2026
AIE  Braintrust  WorkOS  OpenAI

## Full transcript

[0:00](https://www.youtube.com/watch?v=btxGmN8RvNU&t=0s) [upbeat music] Okay, so let's just work with it like this, a small room.

[0:19](https://www.youtube.com/watch?v=btxGmN8RvNU&t=19s) Hi, everybody. Welcome. My name is Rafael, um, representing Bright Data. Uh, Bright Data is basically, uh, web access platform to help agents or anybody collect public data on scale.

[0:31](https://www.youtube.com/watch?v=btxGmN8RvNU&t=31s) And, um, I'm here to talk about the LLMs misleading people and all the time convincing them, "Hey, I did a search. Hey, I did this," uh, while it didn't.

[0:40](https://www.youtube.com/watch?v=btxGmN8RvNU&t=40s) Why? Because LLMs are programmed to please people, please users, so they making up things, and this is the biggest issue right now I'm seeing with LLMs, and I'm building applications all the time.

[0:50](https://www.youtube.com/watch?v=btxGmN8RvNU&t=50s) I would rather LLM tell me, "No, I can't," but it never does. It always tries to make things up. So, um, currently, the web is actually fighting the robots and automations, and it's been something that's gone on for years.

[1:02](https://www.youtube.com/watch?v=btxGmN8RvNU&t=62s) Everybody knows CAPTCHAs. The first CAPTCHAs showed up 10 years ago, like, you know, a decade ago, and it just keeps growing. And now you have AI blocking AI, and there's a whole war going on, and the web access is actually not as simple as it looks.

[1:17](https://www.youtube.com/watch?v=btxGmN8RvNU&t=77s) So they're getting CAPTCHAs, and, uh, they don't actually report CAPTCHAs. So it tries to go and find the data a different way. Sometimes it goes into, uh, training data, and this is where the worst thing is, when it uses training data and tells you that this is the current situation.

[1:34](https://www.youtube.com/watch?v=btxGmN8RvNU&t=94s) Training data is from twenty twenty-four, we're in twenty twenty-six, and it doesn't add up, right? So these are some of the new things that I was just literally checking out, right?

[1:44](https://www.youtube.com/watch?v=btxGmN8RvNU&t=104s) So Cloudflare blocks AI crawling for about twenty percent of the web, right? So twenty percent of the web is literally not accessible by AI, by default fetch that's built into it.

[1:53](https://www.youtube.com/watch?v=btxGmN8RvNU&t=113s) And Cloudflare also right now released an AI Labyrinth to actually trap bots and mislead them and provide to them fake data, so then your results are getting even worse.

[2:04](https://www.youtube.com/watch?v=btxGmN8RvNU&t=124s) Okay? So the invisible failure route, right? There's no error, no warning, just the wrong answer, right? So the agent sends the request, it gets a CAPTCHA. Uh, even an empty page, it doesn't tell you, "Hey, I got an empty page," right?

[2:19](https://www.youtube.com/watch?v=btxGmN8RvNU&t=139s) It will try to make up something, and this is where the most of hallucinations come from, the need to please and lack of data. So it literally makes things up.

[2:27](https://www.youtube.com/watch?v=btxGmN8RvNU&t=147s) I've seen it literally make numbers up, provide fake citations. You click on the citation, it's a 404, the page doesn't exist, and I'm sure all of you have, uh, seen that happening recently.

[2:38](https://www.youtube.com/watch?v=btxGmN8RvNU&t=158s) I mean, literally, like sixty percent of citations on ChatGPT is not working. How many of you have tried to pro- purchase a product, "Hey, find me this product online, I wanna buy it, and give me a link to the product."

[2:48](https://www.youtube.com/watch?v=btxGmN8RvNU&t=168s) You click on the link to the product and there's no product. Like, so what is this product that you're talking about? The URL doesn't exist. The product doesn't exist.

[2:56](https://www.youtube.com/watch?v=btxGmN8RvNU&t=176s) So where do I buy this product for the fifty bucks? It doesn't exist.

[3:01](https://www.youtube.com/watch?v=btxGmN8RvNU&t=181s) Um, so what I wanna show you, and I'm gonna show you in code, how many of you are familiar with coding, like, um, VS Code? No.

[3:09](https://www.youtube.com/watch?v=btxGmN8RvNU&t=189s) Okay, perfect. So nobody's gonna get lost if I'm gonna switch to that. So I'm gonna do a demo basically with MCP and without the Bright Data MCP, and we're gonna compare what is going on there.

[3:22](https://www.youtube.com/watch?v=btxGmN8RvNU&t=202s) So first I'm gonna-- I, I, I wanna show you that I have exactly identical prompts for both of the, uh, scripts, right? So without MCP and with MCP. So the, the...

[3:33](https://www.youtube.com/watch?v=btxGmN8RvNU&t=213s) I, I give it five tasks. Uh, property, rightmove.co. Let's go check out some of the properties. LinkedIn, Instagram, Amazon, and TikTok. These are the five basic, um,

[3:44](https://www.youtube.com/watch?v=btxGmN8RvNU&t=224s) uh, sites that I wanted to access. They are very heavy on anti-bot systems, and I wanna show you the difference. So first I'm gonna run it without an MCP, and I'm gonna literally to, to, let the AI talk f- for itself.

[3:56](https://www.youtube.com/watch?v=btxGmN8RvNU&t=236s) And GPT-5's a bit slow, so it's gonna take some time. But basically, what we're trying to do is we're trying to access this URL. It's, uh, some local properties that I did literally half an hour ago.

[4:08](https://www.youtube.com/watch?v=btxGmN8RvNU&t=248s) Um, LinkedIn company Mistral. Let's check it out. Instagram account, uh, some Amazon product, and some TikTok.

[4:17](https://www.youtube.com/watch?v=btxGmN8RvNU&t=257s) This is not limited to just w- these five websites. It's just something that I picked that is-- I know for sure will not work without MCP. So as you can see, without MCP, I don't have live web data access.

[4:31](https://www.youtube.com/watch?v=btxGmN8RvNU&t=271s) Uh, it doesn't have any browsing tools, right? So this is with tools not available, just by default, out of the box, um,

[4:38](https://www.youtube.com/watch?v=btxGmN8RvNU&t=278s) um, GPT-5. I mean, it's a strong bo-- uh, it's a strong, uh, LLM, and zero success, five failed. Same exact thing, exactly the same prompt,

[4:52](https://www.youtube.com/watch?v=btxGmN8RvNU&t=292s) I'm running with our MCP. Uh, our MCP has sixty-nine tools. While it's running, I just wanna go over some-- the tools that it has. Search engine. Search engine is basically the, uh, LLM is able to do Google search, Bing search, DuckDuckGo search.

[5:09](https://www.youtube.com/watch?v=btxGmN8RvNU&t=309s) Real searches, not just like, you know, search the, the web that it's di- doing in the background. Um, it has...

[5:21](https://www.youtube.com/watch?v=btxGmN8RvNU&t=321s) It also has scrape as a markdown. That's a very strong one. Basically, it can send a curl to any URL and get just the markdown without the HTML tags.

[5:30](https://www.youtube.com/watch?v=btxGmN8RvNU&t=330s) So you're not wasting tokens on parsing the HTML.

[5:34](https://www.youtube.com/watch?v=btxGmN8RvNU&t=334s) Uh, search engine batch, if you wanna do like, you know, like a hundred keywords, you can literally se- it can literally send a hundred keywords and get the hundred keyword back results like, so the scaling is also huge.

[5:44](https://www.youtube.com/watch?v=btxGmN8RvNU&t=344s) Discover. It also has pre-built APIs for many websites, as you can see. And of course, it has a scraping browser infrastructure. So it's a remote browser that your LLM can open and navigate.

[5:55](https://www.youtube.com/watch?v=btxGmN8RvNU&t=355s) The remote browser solves CAPTCHA by itself. It has a unique fingerprint. So it can open a hundred browsers, navigate the same website without getting blocked. So the whole idea is that with our MCP, not only can it do single sessions, it can do multiple sessions in parallel.

[6:11](https://www.youtube.com/watch?v=btxGmN8RvNU&t=371s) And as you can see, let's see, we have success for the Rightmove, we have success for the LinkedIn Mistral, uh, Instagram also worked. Uh, Amazon product, it has the information for the product itself.

[6:26](https://www.youtube.com/watch?v=btxGmN8RvNU&t=386s) Uh, and then what I did is on the second part, I asked the LLM to compare the results from no MCP with MCP, so that you don't take my word for it, let's see what the ChatGPT will actually say,

[6:39](https://www.youtube.com/watch?v=btxGmN8RvNU&t=399s) if it didn't get stuck. It looks like it's stuck for some reason.

[6:49](https://www.youtube.com/watch?v=btxGmN8RvNU&t=409s) May I ask a question?

[6:49](https://www.youtube.com/watch?v=btxGmN8RvNU&t=409s) Of course. Please, I, I love questions.

[6:52](https://www.youtube.com/watch?v=btxGmN8RvNU&t=412s) For social profiles, do you have, like, uh, accounts that you run?

[6:57](https://www.youtube.com/watch?v=btxGmN8RvNU&t=417s) No. We only work with public data.

[7:00](https://www.youtube.com/watch?v=btxGmN8RvNU&t=420s) Public data.

[7:00](https://www.youtube.com/watch?v=btxGmN8RvNU&t=420s) Only publicly available data. Um, collecting data behind login is not really legal. Why? Because when you s- sign up and create an account, you accept terms and conditions. When you accept terms and conditions, you need to really check it if it says, "Can you scrape?"

[7:14](https://www.youtube.com/watch?v=btxGmN8RvNU&t=434s) Can you... Do you allow... Do they allow robots to access the website? And that's why m- maybe some of you heard there's a lot of lawsuits going on, LinkedIn suing these people.

[7:23](https://www.youtube.com/watch?v=btxGmN8RvNU&t=443s) Everybody's suing each other.

[7:24](https://www.youtube.com/watch?v=btxGmN8RvNU&t=444s) Amazon.

[7:25](https://www.youtube.com/watch?v=btxGmN8RvNU&t=445s) Amazon suing, yes. But I think even, yeah, for example, LinkedIn and Instagram, I think they don't even show you really any public data, even if you're on the- There's plenty of public data for, for LinkedIn, of course.

[7:37](https://www.youtube.com/watch?v=btxGmN8RvNU&t=457s) If you take, for example, I will take this URL. I might get blocked. I don't know how is the local IP, but, um, and I open an incognito window, right?

[7:47](https://www.youtube.com/watch?v=btxGmN8RvNU&t=467s) I mean, usually it just shows me, like, really- There you go. A few. So this is a public data that can be collected.

[7:56](https://www.youtube.com/watch?v=btxGmN8RvNU&t=476s) So if you put a person instead of a company name-

[8:00](https://www.youtube.com/watch?v=btxGmN8RvNU&t=480s) Company, it's the same thing. It's the same thing. The only thing is they're very critical, right? So if you are using, let's say, a Wi-Fi, uh, IP of a big event like this, it probably will block you because it's a data center IP, it's low-quality IP.

[8:12](https://www.youtube.com/watch?v=btxGmN8RvNU&t=492s) From home, you could probably access maybe 5, 10 profiles, but eventually it will also ask you to log in. Uh, so we only deal with public data. So, uh, um, when you guys are using us, you are safe in the sense of nobody's gonna come knocking on your doors to sue you. [laughs]

[8:28](https://www.youtube.com/watch?v=btxGmN8RvNU&t=508s) If that makes sense.

[8:29](https://www.youtube.com/watch?v=btxGmN8RvNU&t=509s) Can we, can we provide credentials to our own social accounts?

[8:31](https://www.youtube.com/watch?v=btxGmN8RvNU&t=511s) No. We... Again, we don't deal with data behind login, uh, we consider it to be illegal, so, um, we don't deal with accepting terms and conditions and data behind login.

[8:41](https://www.youtube.com/watch?v=btxGmN8RvNU&t=521s) Only publicly available.

[8:43](https://www.youtube.com/watch?v=btxGmN8RvNU&t=523s) So I imagine you would cache some of these so that you don't have to constantly go and request it. Do you have an idea of how-

[8:49](https://www.youtube.com/watch?v=btxGmN8RvNU&t=529s) We have a whole data set. So if you guys don't want to do live data, and you don't care if it's a few months old, we have data sets literally that you can just filter by, let's say that you're looking...

[8:59](https://www.youtube.com/watch?v=btxGmN8RvNU&t=539s) If we're talking about LinkedIn people, you're looking for AI engineers in a certain area, you can filter it and just get the data set right away.

[9:06](https://www.youtube.com/watch?v=btxGmN8RvNU&t=546s) Okay.

[9:06](https://www.youtube.com/watch?v=btxGmN8RvNU&t=546s) So, and, and your agent will actually have access to that. So your agent can filter the data set and get you the data if we want to, right? So it has access to all the tools.

[9:16](https://www.youtube.com/watch?v=btxGmN8RvNU&t=556s) So here's basically head-to-head comparison by the LLM itself. Without MCP, failed, no live web access. With, it's listed, failed, successful, failed, successful, so that's basically it. Uh, anti-bot bypass, CAPTCHA solving.

[9:30](https://www.youtube.com/watch?v=btxGmN8RvNU&t=570s) Right, so again, our system automatically solves CAPTCHAs, so if your bot navigates to a website with a browser and it has a CAPTCHA, our s- our browser has built-in CAPTCHA-solving solution, so it will automatically solve the CAPTCHA, and your bot can continue browsing without getting blocked.

[9:44](https://www.youtube.com/watch?v=btxGmN8RvNU&t=584s) How much time I got? I don't know. Um, so just to kind of summarize it, right?

[9:51](https://www.youtube.com/watch?v=btxGmN8RvNU&t=591s) This is the biggest hallucinations that you guys see. The agent gets blocked, it needs to please you, and it makes things up. And, um,

[10:01](https://www.youtube.com/watch?v=btxGmN8RvNU&t=601s) fake content also, right? So now, uh, if you wanna Google what is Cl- uh, Cloudflare AI Labyrinth, it's basically a system, once it detects a bot, it's not... doesn't block it, it literally feeds it fake data.

[10:14](https://www.youtube.com/watch?v=btxGmN8RvNU&t=614s) So bigger hallucinations, right? And, um, the easiest fix for this is just to make sure that your agent doesn't get blocked, and it's as easy as to implement our MCP.

[10:26](https://www.youtube.com/watch?v=btxGmN8RvNU&t=626s) Um, our MCP has a free tier of 5,000 requests. So if you guys wanna try it out, um, this is... You can connect with me on LinkedIn if you want to.

[10:35](https://www.youtube.com/watch?v=btxGmN8RvNU&t=635s) Or is this the... Hold on a second. Is this the sign-up for the MCP? I'm lost in these QR codes.

[10:42](https://www.youtube.com/watch?v=btxGmN8RvNU&t=642s) One second. Internet. Yes, please.

[10:51](https://www.youtube.com/watch?v=btxGmN8RvNU&t=651s) Uh, how does it detect if, if there's Cloudflare Labyrinth or not?

[10:55](https://www.youtube.com/watch?v=btxGmN8RvNU&t=655s) Um, so the way we approach it is that we make your agent look like a human being, literally like, uh, there's mouse movement pre-recorded, there's typing. When it types, it's like it's, uh, mimics a real human behavior.

[11:08](https://www.youtube.com/watch?v=btxGmN8RvNU&t=668s) So, uh, the Cloudflare literally just doesn't even ask, are you a robot or not, right? So this is our, our approach. Instead of trying to s- understand how they detect, we make the agent look as human as possible so that it doesn't trigger the actual, uh, blockage.

[11:25](https://www.youtube.com/watch?v=btxGmN8RvNU&t=685s) Misleading data is one of the toughest things that you can actually encounter. A lot of websites right now, and Asia is doing that, right? Hotels, they're literally providing different prices.

[11:34](https://www.youtube.com/watch?v=btxGmN8RvNU&t=694s) You go check out on your phone, you get one price. You can check from your computer, you get a different price. You connect through a proxy, you get a third price.

[11:41](https://www.youtube.com/watch?v=btxGmN8RvNU&t=701s) Which one is correct? It's really hard to tell. When it comes into the domain of misleading, uh, the best bet is to make sure that you, uh, y- your agent looks like a human, and hope for the best.

[11:52](https://www.youtube.com/watch?v=btxGmN8RvNU&t=712s) That, that's basically what the approach is right now. Um, AI Labyrinth was literally released a few... like, a month ago. Uh, I don't have much of statistics on what is, like, you know, exactly how it's working.

[12:06](https://www.youtube.com/watch?v=btxGmN8RvNU&t=726s) All I know is that- It didn't really affect us. We don't see any kind of change in data, and we are collecting petabytes of data on the daily. Like, we have so many customers always scraping.

[12:15](https://www.youtube.com/watch?v=btxGmN8RvNU&t=735s) We're caching data, so we're always comparing the results. We don't see any degradation in the results, uh, so I think we're doing a good job in that case. Uh, this is a QR code that you guys can sign up for.

[12:26](https://www.youtube.com/watch?v=btxGmN8RvNU&t=746s) Uh, we have a GitHub page, uh, as well, uh, githubbrightdata.com,

[12:34](https://www.youtube.com/watch?v=btxGmN8RvNU&t=754s) or githubbrightdata. And another thing that I would recommend for you guys to check out is the Skills page, right? So what we did is we created skills, and I'm gonna have another session in a couple of hours if you're interested in seeing what the Skills does, is basically you can take any agent, tell it to go here,

[12:54](https://www.youtube.com/watch?v=btxGmN8RvNU&t=774s) and w- this page will teach it on how to build a scraper or how to build a pipeline that will collect you the data. It has all the information it needs, all the APIs, and, uh, if you will come back to my next session, which is at 1:00, I think, something like that, I'm gonna literally demonstrate how

[13:15](https://www.youtube.com/watch?v=btxGmN8RvNU&t=795s) it builds the pipeline. Literally, in front of you, I'm gonna tell it, "Hey, listen, let's build a Walmart collector for ABC," and it's gonna build it, and it's gonna scrape it, and instead of parsing each individual HTML, it's gonna build a parser, and it saves about 99% of the tokens.

[13:28](https://www.youtube.com/watch?v=btxGmN8RvNU&t=808s) 'Cause I see a lot of people as like, "Hey, I need to parse 10,000 pages, but it's so token heavy." Don't parse with the LLM. LLM builds the parser, and then the script runs it, but that's the next session.

[13:40](https://www.youtube.com/watch?v=btxGmN8RvNU&t=820s) Any questions?

[13:41](https://www.youtube.com/watch?v=btxGmN8RvNU&t=821s) Um, question on the performance. Uh, I saw you MCP exposes 69 tools, which means if I s- if I need, let's say, search capability for my information, do I need to load all the 69 tools?

[13:52](https://www.youtube.com/watch?v=btxGmN8RvNU&t=832s) No. Of course, filter it. Yeah, I just showed 69 tools because just to show it. Uh, if I need just a scrape markdown, a search, I would just literally load two tools.

[14:01](https://www.youtube.com/watch?v=btxGmN8RvNU&t=841s) Okay.

[14:01](https://www.youtube.com/watch?v=btxGmN8RvNU&t=841s) Uh, otherwise, you're flooding context with irrelevant data. Of course not.

[14:07](https://www.youtube.com/watch?v=btxGmN8RvNU&t=847s) The, uh, experiment at the beginning. Does that use the web search tool?

[14:12](https://www.youtube.com/watch?v=btxGmN8RvNU&t=852s) I'm sorry. I didn't hear the beginning. Again?

[14:14](https://www.youtube.com/watch?v=btxGmN8RvNU&t=854s) Uh, the, the experiment at the beginning, uh, like your MCP

[14:18](https://www.youtube.com/watch?v=btxGmN8RvNU&t=858s) Yes.

[14:18](https://www.youtube.com/watch?v=btxGmN8RvNU&t=858s) ... used the web, web search tool through the OpenAI API, or is that just like a call to Bright Data? I'm sorry, like, how does that compare? I've not had experience with doing this before, but

[14:31](https://www.youtube.com/watch?v=btxGmN8RvNU&t=871s) Um, so we didn't do any searches. Um, I literally told it, "Hey, go to this URL. See if you can load it," right? So the-

[14:38](https://www.youtube.com/watch?v=btxGmN8RvNU&t=878s) Sorry.

[14:38](https://www.youtube.com/watch?v=btxGmN8RvNU&t=878s) I didn't use the search, uh, in this demo.

[14:41](https://www.youtube.com/watch?v=btxGmN8RvNU&t=881s) Sure.

[14:42](https://www.youtube.com/watch?v=btxGmN8RvNU&t=882s) Uh, but of course, again, even with our MCP, it can actually do a Google search.

[14:47](https://www.youtube.com/watch?v=btxGmN8RvNU&t=887s) Yeah.

[14:48](https://www.youtube.com/watch?v=btxGmN8RvNU&t=888s) And that's the... One of the biggest benefits is because we're used to Google results. So by default, when you're asking LLM, you expect it to do a Google search, but it doesn't.

[14:58](https://www.youtube.com/watch?v=btxGmN8RvNU&t=898s) So the results with, uh, MCP much, much better, and I, I recommend, sign up, try it out. It's free. See the results, compare what you guys get.

[15:07](https://www.youtube.com/watch?v=btxGmN8RvNU&t=907s) Is that 5,000 requests per day or?

[15:09](https://www.youtube.com/watch?v=btxGmN8RvNU&t=909s) Per month. Per month, which is, you know, like, for an MVP, for a little experiment, that's more than enough. Uh, we also have pay-as-you-go, so if you do need a little more, it's, it's nothing like, you know, it's...

[15:20](https://www.youtube.com/watch?v=btxGmN8RvNU&t=920s) Okay. [laughs] I mean, just for a prototype, just for banging around.

[15:23](https://www.youtube.com/watch?v=btxGmN8RvNU&t=923s) Yeah, yeah, it's for prototype, it's perfect. I always... I do a lot of hackathons, and I'm always recommending, "Hey, listen, set up an MCP and tell your agent to go build whatever you need."

[15:31](https://www.youtube.com/watch?v=btxGmN8RvNU&t=931s) It does a much better job than without. [outro music]
