# Revenue Engineering: How to Price (and Reprice) Your AI Product

Kshitij Grover

AI Engineer World's Fair 2025

[Talk page](https://ai.engineer/talks/1C3sZbaxOmw-revenue-engineering-price-reprice-your-ai-product) · [Video](https://www.youtube.com/watch?v=1C3sZbaxOmw)

## Revenue Engineering: How to Price and Reprice an AI Product

Agent pricing connects opaque work to a visible bill. Choosing the right unit means understanding the buyer, the architecture, and how both change as the product improves.

### What should an agent charge for?

[0:28](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=28s)

When an agent does work on your behalf, what should determine the bill? **Pricing is a form of product friction**: it can establish that the product delivers something worth paying for, or prevent someone from discovering that value at all. Where to introduce that friction depends on the audience, the value delivered, and how the product is evolving. Kshitij Grover approaches this problem as a co-founder of Orb, which provides usage-based billing infrastructure across layers of the software stack. In its customer conversations, deciding how to monetize increasingly comes before deciding how to implement billing.

Replit Agent makes the problem concrete. Grover opens with a post from Replit founder and CEO Amjad Masad; Replit is an Orb customer. An agent that builds full-stack web applications could charge a fixed amount whenever it creates a checkpoint, or vary the charge with the complexity of the change. The post contrasts a simple UI change with a large refactor. A checkpoint is an understandable unit, but it does not necessarily represent a consistent amount of work.

![Slide containing Amjad Masad’s post about checkpoint pricing, contrasting a simple UI change with a large refactor.](https://ai.engineer/api/talk-essay-media/20109f46d6322c188533035acc1c89887464993f9d4cf604ec4ed4f7bd2c3b24)

Amjad Masad asks whether Replit Agent checkpoints should have fixed or variable prices.

Choosing between those approaches requires knowing who uses Replit and what costs Replit incurs. It also requires considering the moment the user sees the charge. Someone supplies an initial prompt, but may have little visibility into or control over the subsequent work. A bill presented only after completion can therefore surprise them. That opacity is especially pronounced with agents, although the same pricing questions reach down into AI infrastructure.

### From licenses to variable AI costs

[2:27](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=147s)

Software pricing has moved from buying a fixed TurboTax license at Fry’s Electronics to subscriptions, usage charges, and potentially charges for outcomes.

| Model | What the customer buys |
|---|---|
| Fixed license | Access to a software product |
| Seat subscription | Access for each licensed user |
| Usage-based pricing | Metered consumption |
| Outcome-based pricing | A defined result |

Grover uses a $30-per-seat subscription as an illustration of the familiar middle step. Across these models, traditional pricing aims to remain simple enough to understand, introduce willingness to pay as a signal of value, and protect margins.

Grover cites 80% as a healthy traditional software margin. Infrastructure and AI make margins more variable, though not necessarily unhealthy: an agent can encounter extreme workloads that create disproportionate costs. Predictability consequently matters on both sides of the transaction. A mature customer needs to budget not just for one developer, but for how spending will scale through adoption across the company.

Early AI products face another constraint: buyers may not yet understand what the product can do. They need to experiment and see proof before committing money. Moving the payment decision later can help establish that proof, even while the provider incurs variable costs of goods sold, or **COGS**. Those costs also move quickly. Grover uses an unspecified OpenAI model-cost cut of one-third overnight to illustrate how abruptly an agent’s underlying economics can change. The point is the speed of change, rather than a stable cost assumption on which to build a pricing model.

### Price for the buying journey

[4:43](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=283s)

Three considerations organize the pricing decision: **audience and value delivery**, **margin structure**, and **flexibility to experiment**. Margin structure means understanding the axes along which costs scale, rather than optimizing only the literal margin on day one. That distinction matters particularly for an early product whose underlying costs are changing rapidly. Flexibility matters because the first pricing model is unlikely to be the final one.

Audience determines what a sensible purchasing experience looks like. ServiceNow’s contact-sales pricing page can seem unnecessarily obstructive to a developer, yet fit an enterprise buyer with an existing commitment and an opportunity to buy additional products. The conversation with sales helps tailor the purchase to that relationship. An individual developer, by contrast, may have no procurement team or formal purchasing process and expect to start trying the product immediately.

Look beyond whoever first opens the pricing page. Can that person introduce a vendor on day one, or does adoption require a cross-functional decision? What do they need to establish first? If the immediate need is proof that the product works, provide that proof before inserting a pricing decision. The right amount of friction depends on the buyer’s next step.

### Packaging tells customers how to use the product

[6:36](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=396s)

Replit’s pricing page makes its tiers transparent and offers a free entry point, letting users discover value before thinking hard about payment. The package then communicates the intended use: which models are available, how many agent checkpoints are included, and how many seats come with the plan. Those choices suggest whether the product is collaborative or individual, and whether it belongs in a daily workflow or occasional use. They shape user incentives while controlling the provider’s backend costs.

![Four-column Replit pricing page with a free Starter tier, highlighted Replit Core plan, Teams plan, and Enterprise custom pricing.](https://ai.engineer/api/talk-essay-media/48d10d0c0f00d8c661bb8b1f6baa28d6deb8e5d46b388ab8490c290bd6068e54)

Replit’s pricing page presents Starter, Replit Core, Teams, and Enterprise tiers.

[Unify](https://www.unifygtm.com/pricing), aimed at revenue and go-to-market teams, provides a different example. Grover asks the audience to notice its higher displayed price before considering whether the charge is monthly or annual: the price itself communicates positioning. In the historical page he discusses, the two higher tiers have custom pricing. The displayed allowances are 50,000, 200,000, and 600,000 credits; Grover explicitly says he does not know what those credits mean. Their apparent scale nevertheless conveys something about the workflows the product expects to support. These are the talk’s historical allowances, not the current page’s packaging.

Customer logos complete that positioning. Developer brands and Fortune 100 enterprises signal different intended buyers. **The pricing experience is larger than the price point**: tiers, allowances, access paths, and customer references together tell visitors whether the product is meant for them.

### Turn architectural advantages into pricing advantages

[8:28](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=508s)

As a product’s architecture changes, its margins change with it. That creates an opportunity to pass engineering improvements through to customers in the pricing model. [Cloudflare Workers](https://blog.cloudflare.com/workers-pricing-scale-to-zero/) illustrates the connection: its isolate-based architecture supports charging for CPU milliseconds. When an agent calls OpenAI or Anthropic and waits for a response, that idle network time does not consume billable Workers CPU time. External model-provider charges and other applicable service fees still apply.

The distinction matters for agents because elapsed time can include substantial waiting on external services.

| Meter | Treatment of an external API wait |
|---|---|
| CPU time | Idle waiting does not consume CPU time |
| Wall time | Waiting remains part of elapsed duration |

An architectural decision therefore becomes a customer-facing advantage for this workload. Grover treats this as a way to pass technical innovation to users through pricing.

Protecting margins does not require charging proportionally for every increment of usage. First identify the extreme workloads that could create unacceptable exposure; then consider how to prevent them or encourage reasonable usage. **Rate limits and guardrails** can protect the business without forcing the customer to experience a continuously rising bill for every action.

### Remove credit counting from the workflow

[10:01](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=601s)

Jasper’s [Brand Voice and Unlimited Credits announcement](https://www.jasper.ai/blog/release-update-brand-voice) connects architecture to a different kind of customer benefit. Grover presents it as a historical example and notes that the post and pricing may have changed since he collected them. Marketing teams writing copy do not want to count credits. Jasper’s architecture could switch between models under the hood, removing the need for users to choose the right model for each job. In the example, Jasper moved to unlimited credits across all three tiers.

![Slide titled “A ‘good-fit’ architecture highlights the core value props of the product,” with overlapping Jasper Brand Voice and Unlimited Credits excerpts.](https://ai.engineer/api/talk-essay-media/722dd3d5f2f58da52e88efcf1073f2b7de42928e1ce154e31870f63de0043a3e)

Jasper’s “Unlimited Credits” announcement accompanies its Brand Voice description.

That packaging supports the product’s intended role as a routine part of marketing work. A customer who fears running out of credits must decide which tasks deserve Jasper; a customer without that recurring decision can use it throughout the workflow. The architectural advantage and the pricing model reinforce the same value proposition, without requiring the user to manage the underlying model choices.

### Repricing changes the organization too

[10:57](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=657s)

Product teams continuously invest in R&D, but traditional seat pricing often captures that added value only periodically. Grover illustrates the mismatch with an annual increase from $20 to $25 per seat. An agent can make incremental value more visible: users take an action, receive an output, and see the quality improve. That closer connection between action and result can make an evolving price easier to understand.

Grover reports that some customers change prices two or three times a month. That cadence creates work beyond editing a price table. Customers must understand the changes, and the business must manage the resulting complexity in their expectations.

Internal incentives also need to follow the pricing model. A commissioned sales team selling large enterprise contracts operates differently from one supporting pay-as-you-go adoption. Grover describes usage-based companies that commission representatives on account expansion over time rather than the initial contract alone. That shifts attention toward what happens after the sale, raising a further question: how should sales and customer success divide responsibility for growing the account? Repricing can change the organization as well as the bill.

### Simulate the bill before changing the price

[12:59](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=779s)

Before applying a new pricing model, simulate its effect on customer usage. At meaningful operating scale, an aggregate revenue estimate is insufficient: customers with different use cases can respond very differently to the same charging rule.

1. Group customers into cohorts with distinct usage patterns.
2. Calculate what those patterns would cost under the proposed model.
3. Inspect how the revenue mix changes across cohorts.
4. Examine the highest-revenue customer individually to understand its exposure.

This makes repricing a decision informed by observed customer behavior.

A small TypeScript calculation makes that comparison concrete. Here, constructed usage records compare a fixed checkpoint charge with a proposed complexity-weighted charge. The rates and complexity values are teaching inputs, not Replit’s pricing. Both calculations use the same usage records; the proposal changes only the charging rule.

```typescript
type Usage = {
  customer: string;
  cohort: string;
  checkpoints: number;
  complexityUnits: number;
};

const usage: Usage[] = [
  { customer: "a", cohort: "small-edits", checkpoints: 100, complexityUnits: 120 },
  { customer: "b", cohort: "large-changes", checkpoints: 40, complexityUnits: 240 },
];

const currentBill = (u: Usage) => u.checkpoints * 20;
const proposedBill = (u: Usage) => u.complexityUnits * 10;

const comparison = usage.map(u => ({
  ...u,
  currentCents: currentBill(u),
  proposedCents: proposedBill(u),
  deltaCents: proposedBill(u) - currentBill(u),
}));

const cohorts = [...new Set(usage.map(u => u.cohort))].map(cohort => {
  const rows = comparison.filter(row => row.cohort === cohort);
  return {
    cohort,
    currentCents: rows.reduce((sum, row) => sum + row.currentCents, 0),
    proposedCents: rows.reduce((sum, row) => sum + row.proposedCents, 0),
  };
});

const topCurrentCustomer = [...comparison]
  .sort((a, b) => b.currentCents - a.currentCents)[0];

console.table(comparison);
console.table(cohorts);
console.log({ topCurrentCustomer });
```

The comparison separates customer impact from total revenue impact. It estimates bills under unchanged usage; it does not predict how customers would alter their behavior after repricing.

### From unlimited plans to choosing a spending plan

[13:34](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=814s)

Grover closes with predictions for AI agent pricing in 2025. He expects continuing price competition and cheaper agents to push providers toward effectively unlimited plans. Those plans would still have caps and guardrails, but their aim would be to make agents habitual participants in every workflow rather than tools customers reserve for selected tasks.

He also expects outcome-based pricing, already used by some companies, to become more concrete. Its difficult part is defining the **service-level agreement**, or SLA: what counts as the outcome, how it is measured, and whether that definition makes sense to the customer. Charging for success requires contracts precise enough that both parties can recognize success when it occurs.

The final prediction returns to the opening problem of an opaque agent producing a surprising bill. Real-time spending visibility, balance alerts, and spend management could give users more control. Grover compares a possible agent experience to a database query planner: before executing a prompt, the agent could estimate consumption or offer alternative execution plans. His hypothetical includes three options, with illustrative estimates of 10,000 and 4,000 credits for two of them. Those are proposed costs before execution, not measured consumption. He expects this kind of customer control to become more sophisticated within the year of the talk.

![Blue “AI Agent Pricing in 2025” slide with three numbered columns discussing price wars and COGS pressure, outcome SLAs, and sophisticated customer controls.](https://ai.engineer/api/talk-essay-media/eb50a5dde44687eec78bb6bb9b9dee3feda966f7ee624a09e10e6231a22f49e2)

Three AI agent pricing predictions cover cost pressure, outcomes, and customer controls.

For further tactical detail, Grover points to Orb’s *Pricing AI Agents* e-book, listed in its [e-book catalog](https://www.withorb.com/blog), and names threshold billing, fraud, and prepaid pricing models as topics to explore there. Those implementation concerns sit beyond this talk’s central design problem: helping customers understand and control what they will pay as agents do more work on their behalf.

## Resources

- [Workers pricing and idle I/O](https://blog.cloudflare.com/workers-pricing-scale-to-zero/) — Cloudflare explains CPU-time billing and why waiting for network responses does not consume billable CPU time.
- [Jasper's move to unlimited credits](https://www.jasper.ai/blog/release-update-brand-voice) — Jasper's historical announcement explains its three paid tiers, removal of credit limits and automatic model selection.
- [Replit introduces effort-based Agent pricing](https://replit.com/blog/effort-based-pricing) — The June 2025 announcement explains the transition from fixed-price checkpoints to charges reflecting task effort.

## Slides

### Innovation Partner and Platinum Sponsors

[0:10](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=10s) [AWS is the innovation partner; six platinum sponsor logos appear below.](https://assets.ai.engineer/sites/aiecode2025/sha256/f9bb32b7a9cf9118d07ecc05ff6aae2e4d2726ed1c43036cb1930e8e38517cb6.webp)

INNOVATION PARTNER
aws

PLATINUM SPONSORS
Graphite
Windsurf
MongoDB®
daily™
augment code
WorkOS

### Revenue Engineering

Kshitij Grover frames pricing as friction: it can serve a purpose or keep people from using a product. He ties pricing decisions to the product’s value and audience.

[0:28](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=28s) [Orb title slide: How to price and reprice your AI agent, presented by Kshitij Grover.](https://assets.ai.engineer/sites/aiecode2025/sha256/eee489dfa8877fbd360cb2bb5069c4ba694106b3844e4ddff6033b553a1d031f.webp)

Revenue Engineering
How to price and reprice your AI agent

KSHITIJ GROVER
Co-founder & CTO at Orb

orb.

### Orb Partners Across the AI Stack

Orb’s partners span data retrieval, infrastructure, AI apps, and developer tools. Grover says conversations often start with monetization and pricing before moving into billing implementation.

[1:07](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=67s) [Orb partner logos grouped under data and retrieval, model and GPU infrastructure, AI native apps, and developer tools and agents.](https://assets.ai.engineer/sites/aiecode2025/sha256/a25de50e2c2a8f3adfaeeb993ff0f471e49f213a79fe1c68068bb21a846bec59.webp)

Orb partners with innovators at every layer of the AI stack

Data & retrieval
Chroma
WarpStream by Confluent
ragie.ai
exa
Airbyte
contextual ai
Pinecone
activeloop

Model & GPU infra
Fireworks AI
TensorChord
fal
baseten
AKKA

AI native apps
Assured
Ingredient AI
perplexity
momentic
Reality Defender
maisa
LiquidMetal AI
Inflection AI
Stokk.ai
WALLFACER LABS
klipy
Greeter.AI

Dev tools & agents
ARCADE
Relevance AI
kodey
Eachlabs
Kilo Code
zive.
replit
FACTORY
Lindy
braintrust
augment code

orb.

### In pricing AI every layer of the AI stack

Amjad Masad asks whether Replit Agent should charge a fixed price per checkpoint or vary it with the work done. Variable pricing could better reflect complexity, but opaque agent work can leave users surprised by the bill.

[1:11](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=71s) [A post discusses fixed versus variable checkpoint pricing for Replit Agent.](https://assets.ai.engineer/sites/aiecode2025/sha256/2f076faffc2a764fabdc6f1a81c741bed0407a1795e5ac06539c13fa36707b51.webp)

In pricing AI every layer of the AI stack

Amjad Masad
@amasad

Should Replit Agent move from constant price per checkpoint ($1/4) and go variable price per checkpoint?

In other words, each checkpoint price is proportional to work done. Eg simple UI change could be 2c, while a massive refactor could be 50c.

9:10 AM · May 26, 2025

orb.

[1:15](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=75s) [A slide presents a post asking whether agent checkpoint prices should vary with the work performed.](https://assets.ai.engineer/sites/aiecode2025/sha256/6207209a469ee0bcad7b51c1e66b87cc8034b987735604d144590c7e09e56262.webp)

In pricing AI every layer of the AI stack

Amjad Masad
@amasad

Should Replit Agent move from constant price per checkpoint ($1/4) and go variable price per checkpoint?

In other words, each checkpoint price is proportional to work done. Eg simple UI change could be 2c, while a massive refactor could be 50c.

9:10 AM · May 26, 2025

orb.

Interface text:
Microsoft
smol ai

### Monetization is changing...but some things stay true

AI pricing pairs simplicity with predictable costs and balances entry barriers with repeatable income while making room for experimentation. Margins still matter as COGS are higher and change quickly.

[3:39](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=219s) [Traditional pricing principles of simplicity, entry barriers and margins are paired with AI-native concerns of predictability, speed to value and changing costs.](https://assets.ai.engineer/sites/aiecode2025/sha256/f43eae46eb8a4bf088b05a53e25c512354a6efbbc31612c5798b8c6e05336393.webp)

Monetization is changing...but some things stay true

TRADITIONAL PRICING

Simplicity is key to understanding the full cost/benefit of a tool

Barriers to entry need to be balanced with repeatable income

Margins matter to business success

AI-NATIVE PRICING

Predictability matters for companies dealing in tokens and credits for the first time

Speed to value and experimentation are king

COGS are higher and changing faster than ever

orb.

### Key principles

Price for your audience, track how costs scale even before charging for AI, and leave room to experiment with pricing and packaging as costs change.

[4:42](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=282s) [Three columns recommend considering the audience, tracking margin structure and keeping pricing and packaging flexible.](https://assets.ai.engineer/sites/aiecode2025/sha256/c8cc6eb13033558513438d5dda9e6df1c350e59a6de60c93a56774b82536f40d.webp)

Key principles

Audience
Decide who or what the pricing is optimized for – and be mindful to how it plays with the incentives of your audience.

Margin structure
Track costs, even if you’re not directly charging for AI features (yet).

Flexibility
Give yourself the ability to experiment with pricing & packaging over time – you won’t get it right the first time!

orb.

Microsoft
ai

[4:46](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=286s) [Three pricing principles: audience, margin structure, and flexibility.](https://assets.ai.engineer/sites/aiecode2025/sha256/bf4d1e4b455f4a0c7be46a9c1681573386482024765286b1c8dc44c9114555b0.webp)

Key principles

Audience
Decide who or what the pricing is optimized for – and be mindful to how it plays with the incentives of your audience.

Margin structure
Track costs, even if you’re not directly charging for AI features (yet).

Flexibility
Give yourself the ability to experiment with pricing & packaging over time – you won’t get it right the first time!

orb.

### Pricing for your audience

Enterprise deals may involve sales and a buying process, while individual developers may want to try immediately. Consider whether cautious users need proof before pricing.

[5:22](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=322s) [Audience pricing considerations: barriers to entry, purchasing practices, and adoption curve.](https://assets.ai.engineer/sites/aiecode2025/sha256/df50942ae3fc6d15337e19fbb998c30de3e57e9fbfa6bfa359284ca2a1a0be89.webp)

AUDIENCE
Pricing for your audience

COMPANY SIZE
Barriers to entry
Prepaid commitments vs PAYGO influence a company’s willingness to roll the dice

BUYING GROUP
Ways of purchasing
How do they normally like to buy? What pricing models are familiar to them?

ROLES
Adoption Curve
Are you selling to early adopters? To cautious users seeking reliable proof points?

orb.

### Replit pricing example

Replit’s free tier lets users see value before paying. Its paid tiers package model access, agent checkpoints and seats to signal different uses of the product.

[6:36](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=396s) [Replit pricing page comparing Starter, Core, Teams, and Enterprise plans.](https://assets.ai.engineer/sites/aiecode2025/sha256/ec7b1a885cead790ae11a59139fac907135cd1bbe4cc2bf4d7e33cb9302230f0.webp)

Pricing

Monthly
Yearly
Save $60

Starter
Free
Explore the possibilities of making apps on Replit.
Sign up
✓ Replit Agent trial included
✓ 10 development apps (with temporary links)
✓ Public apps only

Replit Core
$20 per month billed annually
Make, launch, and scale your apps.
Join Replit Core
✓ Full Replit Agent access
✓ $25 of monthly credits (~100 Agent checkpoints)
✓ Unlimited public and private apps
✓ Access to Claude Sonnet 4 & OpenAI GPT-4o
✓ Deploy and host live apps
✓ Pay-as-you-go for additional usage
Want to sponsor multiple seats for students or community developers? Contact Sales

Teams
$35 per user per month billed annually
Bring the power of Replit to your entire team.
Join Replit Teams
← Everything included with Replit Core
✓ $40/mo in usage credits included
✓ Credits granted upfront on annual plan
✓ 50 Viewer seats
✓ Centralized billing
✓ Role-based access control
✓ Private deployments
✓ Pay-as-you-go for additional usage

Enterprise
Custom pricing
Meet your security and performance needs.
Contact us
← Everything in Teams
✓ Custom Viewer Seats
✓ SSO/SAML
✓ SCIM
✓ Advanced privacy controls
✓ Custom pricing
✓ Dedicated support
Enterprise plan requires annual commitment and a minimum of 20 users. Contact Sales

Interface text:
replit
Features
Teams
Pricing
Guides
Blog
Careers
Log in
Start building

### Unify pricing tiers

Unify’s Growth tier starts at $1,460 per month billed annually; Pro and Enterprise use custom pricing. The speaker says the large credit allowances and customer logos also shape visitors’ expectations.

[7:30](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=450s) [Unify pricing page comparing Growth, Pro and Enterprise tiers, including their credit allowances.](https://assets.ai.engineer/sites/aiecode2025/sha256/80449366179511a2da31399db646b0c5247f71c53393fbf94b975c83db45e64c.webp)

[illegible]erplexity  CURSOR  MERGE  flock safety  together.ai  hightouch  SoFi  GURU

Use our pricing calculator →
Most popular

Growth
Starting from
$1460 per month, billed annually
For teams looking to get started with signal data
What's your work email?
Talk to sales
50,000 Credits
1 Sequencing User
$100 / seat / mo for additional sequencing users
8 Unify Managed Gmail Mailboxes
$25 / mailbox / mo for additional mailboxes
Onboarding + Support

Pro
Custom
For teams looking to start automated outbound
What's your work email?
Talk to sales
200,000 Credits
2 Sequencing Users
$100 / seat / mo for additional sequencing users
20 Unify Managed Gmail Mailboxes
$25 / mailbox / mo for additional mailboxes
Tailored Onboarding + Support

Enterprise
Custom
For teams looking to scale automated outbound
What's your work email?
Talk to sales
600,000 Credits
5 Sequencing Users
$100 / seat / mo for additional sequencing users
40 Unify Managed Gmail Mailboxes
$25 / mailbox / mo for additional mailboxes
White Glove Onboarding + Support

Interface text:
AIE
Microsoft
smol ai

### Consider your margins

Changing models and architecture shift margins. Cloudflare Workers bills for CPU milliseconds instead of wall time; rate limits or guardrails can address extreme workloads.

[8:27](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=507s) [Three margin considerations: model evolution, technical innovation and unpredictable workloads.](https://assets.ai.engineer/sites/aiecode2025/sha256/e4c13e3be46262ddaee8917097ea9c3b004f4c18aa4e31394c9c22f92eed9ec8.webp)

MARGIN STRUCTURE
Consider your margins

Model Evolution
Introduces rapid change, new consumer expectations, and unique COGS

Technical innovation
Introduces greater need for repeatable revenue

Unpredictability
Defending against degenerate workloads and incentivizing reasonable usage

orb.

### A good-fit architecture highlights the product’s core value

Jasper’s example paired automatic model selection with unlimited credits across three tiers, removing credit-counting from marketing workflows. The speaker notes that the post and pricing may have changed.

[10:00](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=600s) [Slide uses Jasper Brand Voice and unlimited credits examples to illustrate a product architecture aligned with customer value.](https://assets.ai.engineer/sites/aiecode2025/sha256/1951fc37a0180ab29688c20eb0beb02f4287835d4df2105f72d1ee0c1ccf485a.webp)

A ‘good-fit’ architecture highlights the core value props of the product

Jasper Brand Voice

Businesses today face impossibly high content demands. AI writing tools help, but outputs often lack authenticity and alignment with the company’s brand. Jasper Brand Voice enables users to build a memory bank that securely and automatically infuses their AI model with company and product facts, brand tones and style guidelines specific to their company. So when a marketing team uses AI to help develop a 200-asset product launch, each piece can be aligned, on-message and on-brand.

Jasper’s AI Engine is unique in that it can pull from a variety of different models – from OpenAI to Cohere to Anthropic – choosing the right model for the job and increasing reliability. From there, Jasper’s engine enriches the AI with the [illegible]

Unlimited Credits

In addition to the above updates, we’ve gone unlimited on each of our three tiers. No more counting credits or predicting word count. We heard from customer feedback that credits caused confusion in the purchase process and inhibited use, so we’ve removed the need to purchase credits from paid plans.

Hopefully these updates add value to your Jasper experience. There are guides inside the app and you can access help documentation here.

[illegible] change your plan, you can visit the billing page. For now, just enjoy this no-extra-charge upgrade on us. The new plans are as follows:

orb.

### Pricing cannot be a set it and forget it exercise

As product value grows, the charts contrast a sudden price change with incremental price evolution. The speaker argues that gradual repricing can track improvements customers see.

[10:59](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=659s) [Two charts contrast a sudden price change with incremental price evolution as product value grows; bullets emphasize changing markets, experimentation and flexibility.](https://assets.ai.engineer/sites/aiecode2025/sha256/52ca46b408e5a3054bf2606045bfd89881986d8cbce9ac4890aaee1571819e6e.webp)

FLEXIBILITY
Pricing cannot be a set it and forget it exercise

VALUE
Product value for customers
Sudden Price Change
TIME

Product value for customers
Incremental price evolution

• Products, COGS and markets change
• Capturing value is an experiment, not a rule
• The limitless flexibility of AI requires flexibility in how we monetize

orb.

### Repricing

Keep new pricing understandable, plan for sales incentives and contract migrations, and simulate how a price change affects customers with different usage patterns.

[12:09](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=729s) [Three repricing considerations: control complexity, manage change and use data to simulate customer impacts.](https://assets.ai.engineer/sites/aiecode2025/sha256/f65aa37d77d65de1a3b856eb0490afbfba57064513cd180e0457d7b1016153e2.webp)

Flexibility
Repricing

Control Complexity
Make new price models simple to understand and help customers project future costs

Manage Change
GTM team commission predictions, contract migrations

Iterate With Care
Use data to simulate the impacts of a change and identify impacted customers.

orb.

### AI Agent Pricing in 2025

The speaker predicts cheaper agents and capped “effectively unlimited” plans, clearer definitions of paid outcomes, and real-time controls over customer spend and credit risk.

[13:32](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=812s) [Three predictions: continuing price wars and cost pressure, outcome-based pricing, and more sophisticated customer controls.](https://assets.ai.engineer/sites/aiecode2025/sha256/18b932f4e7f1eada421b910fce6f16813b02f9a729c97573199307b73f72b0d1.webp)

PREDICTIONS
AI Agent Pricing in 2025

01
The price wars continue, and so does COGS pressure

Competitive markets will lead to races to the bottom – expect more "effectively unlimited" plans with abuse caps

02
Outcomes get real (SLAs) and the stakes go up

We'll see more "clear definitions of success" as companies price on support ticket resolution, commercial GTM wins, or revenue shares.

03
More sophisticated customer controls - get ready for the audit

Customers will need to get real-time visibility and the option to manage credit risk in real-time. No one can afford to not have a throttle option.

orb.

### AI Agent Pricing E-book

Orb’s AI agent pricing e-book offers tactical advice on threshold billing, fraud, and prepaid pricing models.

[15:17](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=917s) [Closing slide displaying a link to Orb’s AI agent pricing e-book.](https://assets.ai.engineer/sites/aiecode2025/sha256/7b3ea3e9210a026e0212bdfee24e7dc7c5119cd6330d5db95874d92068530241.webp)

withorb.com/pricing-ai-agents

orb.

## Full transcript

[0:00](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=0s) [upbeat music] Thanks for coming to my talk.

[0:16](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=16s) I'm Kshitij. Uh, I'm one of the co-founders at Orb, um, and I'm gonna be talking about how to think about pricing. Um, maybe top-level takeaway, uh, from this talk is that pricing is a, is a deep, complicated topic.

[0:28](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=28s) We're gonna cover some examples. We're gonna cover some tactical advice. Um, but in general, the way you should think about pricing is pricing is a form of friction, uh, for your product.

[0:36](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=36s) And sometimes that friction can be applied for a very good reason. Sometimes that friction can just prevent people from using your product. And so you, you really have to think about the value your product is delivering as well as the audience you're delivering it to, um, and the way that you're building and innovating in your product to,

[0:50](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=50s) to start really thinking about pricing. Um, just a quick, uh, intro to Orb. Uh, Orb is a usage-based billing infrastructure company, so we help companies at lots of layers of the stack, um, think about pricing and billing.

[1:03](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=63s) Uh, nowadays, a lot of our conversations first center around monetization and pricing as a practice, and then we get into the weeds of how do we actually bill for this stuff.

[1:12](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=72s) So, uh, you know, maybe, maybe an intro, uh, tweet to start. Um, this is Amjad. Uh, he's the CEO and founder of Replit. Replit is one of Orb's customers.

[1:22](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=82s) Um, and, and this is a tweet, actually pretty recent, starting to think about how Replit Agent should price itself. And so, so, you know, pretty short tweet, but a lot of complexity actually packed in here.

[1:31](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=91s) Uh, really starting to think about, you know, as you have Replit, which is a very popular agent that, you know, does programming, s- uh, is starting to build full stack web apps, um, how sh- how should a company like Replit price their agent?

[1:43](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=103s) Should it be like a s- super simple every time the agent checkpoints, it charges you a fixed amount, um, or should it start aligning with the complexity of the change it made?

[1:53](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=113s) Um, and as you can start, you know, thinking through this question, you have to think through considerations like who is Replit built for, right? What are the costs that Replit incurs?

[2:00](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=120s) Um, but also just the user experience of if I get charged some amount at the end of something that this agent has done without me really having control over what the agent did other than the initial prompt, um, is that gonna be surprising?

[2:12](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=132s) Is that not gonna be surprising, right? So, um, y- these considerations apply particularly to agents because the work that the agent is doing is fairly opaque or can be opaque.

[2:22](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=142s) Um, but they actually apply at every layer of the stack, including the kind of infrastructure layer of the AI stack.

[2:27](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=147s) Um, so monetization as a whole is changing. Uh, maybe let's start with some kind of traditional principles in, in pricing and, and how pricing has always been thought about, right?

[2:35](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=155s) So we're maybe coming from way back in the day, you know, you're, you're selling a fixed license. You, you go to like Fry's Electronics and, and buy your TurboTax license.

[2:45](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=165s) Uh, and then of course, now you have seat-based pricing, subscription, maybe you're paying $30 a seat, um, all the way to you have this usage-based pricing, which is very dynamic, and, and then maybe you can consider outcome-based pricing as a further evolution of that.

[2:57](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=177s) So tr- in traditional pricing, you, you have a fe- a few key principles. You wanna make sure that the pricing is simple so that users can actually understand it.

[3:04](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=184s) They can grok your pricing. They know how they're gonna pay for your tool. Um, you wanna produce some friction because it, it determines the value of your product, like willingness to pay is a very real signal, and then you wanna protect your margins, right?

[3:16](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=196s) Um, traditionally, software margins have been very healthy. Uh, you know, a good software margin looks like 80%. Obviously, as you start getting closer to the infrastructure layer and as you start moving into the world of AI, uh, these margins are, are much, much more variable.

[3:30](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=210s) Not necessarily unhealthy, but, um, there, there's a lot more degenerate workloads that, that you can incur with an agent. So now, now starting to think about AI-native pricing, predictability matters a lot, right?

[3:40](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=220s) I, I think especially when you're selling to more mature companies that need to budget and think about their, uh, you know, cost profile using your tool, they need to be able to predict how that cost is gonna scale, not just on an individual developer, how that cost might scale, but just generally throughout the company's buying process.

[3:59](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=239s) Um, AI is very early, so, uh, speed and, uh, and, and just showing the value of your product and people being able to play around and experiment matters a lot.

[4:09](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=249s) Uh, oftentimes, at least in this space, it's not like people are coming to you buying a very established product. It's kind of your burden of proof to say, "You know, here's how our agent works.

[4:18](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=258s) Just go try it," and then start thinking about pricing. So, so maybe the, the friction point and where it is has changed. And then as I was saying, um, COGS are variable, but, but not only are they higher, they're also changing very, very quickly.

[4:29](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=269s) I'm sure everyone in this room has seen, okay, great, um, you know, OpenAI, uh, cut, cut model costs by a third overnight. Uh, that has a pretty material impact on like potentially the underlying COGS of many of these agentic tools.

[4:43](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=283s) So, um, the three key principles that, that I think matter here are like really thinking about the audience, who your product is optimized for, what the value delivery mechanism looks like, your margin structure.

[4:54](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=294s) Um, and, and I say margin structure and not actually margins because again, the underlying costs are changing very, very rapidly, so you wanna think about what are the axes of scaling rather than what is the literal margin on day one, e- especially if you're an earlier product.

[5:09](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=309s) And then you wanna give yourself the flexibility to experiment over time. This comes with a lot of complexity, but I think it's very important because you're just not gonna get it right the first time.

[5:17](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=317s) Um, so, so these are kind of the, the principles to keep in mind. Maybe going through one by one. Um, the, the way to kind of think about pricing for your audience is, um, you know, h- what is their buying journey, and what are they coming to you to, to buy, and what value are they looking to

[5:33](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=333s) get out of it, right? Uh, if you're-- Like I, I, we, we talked to lots of people who are like, "Oh, you know, ServiceNow's pricing is so silly. You go to their pricing page and it says like contact sales."

[5:43](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=343s) It's like, well, that's, that's a very, very different audience that they're selling to, where they're probably cross-selling some, some existing contract that is a, a big enterprise commit. They want you to talk to sales because they really want to tailor the value to their sales motion, right?

[5:58](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=358s) And then on the other hand, companies that are selling to individual developers, there's not a procurement team. There's no purchasing process. They want you to click-

[6:05](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=365s) Try it and start using the product immediately. So, so, um, don't just think about the kind of point of entry, but think about who's behind that at the company.

[6:14](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=374s) Uh, is it gonna be a cross-functional decision? Is it something that someone can just insert into, you know, a- as, as a vendor on day one, or, or is there gonna be more of a process around it?

[6:23](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=383s) And then also, uh, start to kind of think about, uh, what users are seeking as quickly as possible. If they're seeking a proof point, then perhaps you wanna be able to give them that proof point before you insert pricing into the equation.

[6:36](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=396s) Here's, here's an example, again, using Replit. Um, couple things to note here. Uh, y- you'll see that there's, uh, one, a lot of tiers, and pricing is actually quite transparent.

[6:46](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=406s) Obviously, there's a free tier, right, indicating that they want you to see the value without having to do a ton of work, think about the monetization. Um, and, uh, o- o- one thing that's kind of subtle is oftentimes pricing and packaging really starts to, uh, dictate what sort of use case they are imagining you to use, right?

[7:05](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=425s) Like, um, you know, what models you have access to, how many agent checkpoints you have access to included, how many seats you might have included. Uh, it, it kind of starts positioning your product.

[7:15](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=435s) Is it a multiplayer product? Is it something that people are gonna be using daily, or is it something that you're gonna be using, you know, maybe like five times a month?

[7:22](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=442s) Uh, the way you package it really determines the incentives that you're pushing onto your users, and obviously controlling for the costs that you might pay on the back end.

[7:30](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=450s) Here's another example from Unify, which is, I, I think, a, a agent, agentic company, but it's, like, targeted at revenue and go-to-market teams. One, you'll see that, like, the price point is a lot higher.

[7:40](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=460s) Um, I want you to, like, forget whether it's monthly or annual or any of that. Just, like, you come on this page, the price point is higher. That communicates something psychologically.

[7:48](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=468s) And, and then the, like, second two tiers are custom, right? Uh, um, again, I, and, and maybe the, the third thing to point out here is that 50,000 credits, 200,000 credits, 600,000 credits, I, I actually don't know what that means, but that seems like a large number that communicates something about the workflows they expect to participate in

[8:04](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=484s) and, like, what that might mean for how you use Unify, right? Um, of course, like, it, it maybe goes unsaid, but the set of logos you put on your pricing page is also very important.

[8:13](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=493s) Um, are you tailoring it towards, okay, we, we service the best developer brands, or we service, like, Fortune 100 enterprises. Um, and so pricing is, is not just the price point, but, but of course also the, the whole experience of what someone sees when they land on this page.

[8:28](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=508s) Um, let's just talk a little bit about margins. Um, w- you know, as, as, uh, the, the underlying architecture of your product changes, of course your, your margin's gonna change.

[8:38](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=518s) These inputs are changing very quickly. Um, one, one interesting example here is, uh, you, you... As much as possible, you wanna think about, uh, the differentiation of your product also in R&D innovation and try to pass that on to your users as basically pricing leverage.

[8:54](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=534s) Uh, a, a good example of this is, like, Cloudflare has built their infrastructure on, and their Workers infrastructure on these, like, isolates, right? So, so they're kind of lightweight compute machines, and so they can charge for CPU milliseconds.

[9:06](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=546s) And so if someone is building, like, an agentic app on Cloudflare and it makes, like, a, you know, call out to OpenAI or Anthropic, you're not getting charged for that because Cloudflare is just charging you for CPU milliseconds instead of wall time.

[9:18](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=558s) I think that's a really good example of them, you know, taking an architectural bet, um, which, which might be good for many, many reasons, but it's a really good fit for building AI agents or generally AI software on top of Cloudflare.

[9:29](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=569s) So that's... I see that as an example of, like, passing down technical innovation to the end user. Um, and then the other thing to consider about your margins is you, you don't have to protect them at all costs.

[9:40](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=580s) You just have to think about what are the extreme edge cases, and what are you doing to, like, prevent those outcomes, right? So if you have these potentially really degenerate workloads, uh, how do you incentivize reasonable usage?

[9:51](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=591s) Do you put in rate limits or guardrails? Um, you don't have to kind of linearly necessarily scale cost with usage. You, you can protect your margins potentially in, in different ways.

[10:01](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=601s) Um, th- this is an example from Jasper. I, I think this, this blog post and pricing has changed since, since I pulled this example. But, you, you know, Jasper realized, like, "Hey, we're, we're targeted toward marketing teams.

[10:11](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=611s) Marketing teams don't wanna count credits, like, when they're, you know, bas- basically writing marketing copy." And so, uh, they, they talk a little bit in this blog post about how they've architected a- around being able to, like, switch between models under the hood seamlessly.

[10:25](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=625s) You don't have to choose the right model for the job. And as a result of that, they went unlimited on, uh, each of the three tiers, so you're, you're no longer counting credits as you're using Jasper.

[10:35](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=635s) Now, uh, this kind of highlights the core value prop of the product, which is they wanna be a really critical part of that workflow, and they don't want you to have to, like, pick and choose when you use Jasper because you might have credits or might run out of credits.

[10:48](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=648s) So, uh, again, kind of trying to take, is there, like, an architectural, uh, advantage you have, and how can you pass that on to your pricing model?

[10:57](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=657s) Finally, I wanna talk about flexibility. Um, you know, y- [chuckles] you all, as, as you're working in your teams, uh, you're constantly improving your product. You're putting a lot of R&D into it.

[11:06](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=666s) Um, one thing that is, uh, u- used to be true, especially in the seat-based world, is, like, you'd have R&D over here shipping a ton, um, increasing the product value, and then you'd have, like, a once a year, "Okay, we're changing our seat price from $20 to $25."

[11:20](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=680s) Um, one advantage you have in the AI world is, like, uh, the value is much closer to the end user. It's, it's not necessarily, like, a $20 a seat thing.

[11:27](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=687s) It's, like, someone is literally seeing, "Okay, I'm, I'm taking this action. This agent is giving me this output. Okay, it's getting better, it's getting... it's better, it's getting better."

[11:35](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=695s) So I think not only do you as a business have a responsibility to incrementally evolve your price as, uh, you know, y- you invest R&D, but I think it's, it's honestly a lot more understandable for customers why that's happening because the, the monetization is much closer to the value of the product.

[11:52](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=712s) Um, so, so, uh, you know, obviously, i- in Orb's customer base, but in general, we are seeing people, uh, make many, many more price changes, not just once a year or twice a year, but in fact, some customers making price changes, uh, two, three times a month.

[12:07](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=727s) Um, and again, the- there's, there's caveats that come with that. Uh, you have to think about, you know, what, what is that complexity that introduces in the, in the mental model of your end users?

[12:17](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=737s) How are you gonna manage change? Um, and a, a g- a kind of good internal dynamic example of this is, uh, if your go-to-market team is, you know, commissioned and, and they're selling these big enterprise contracts and your pricing is changing, or you're going to, like, a pay-as-you-go model, you need to think about, like, what does that

[12:32](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=752s) mean for your sales team? What, what does that mean for the incentives that they have in selling your product? Classic example of this is a lot of usage-based companies now commission their sales reps not on the initial contract land, but on the expansion of that- account over time?

[12:46](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=766s) And then, okay, second order effect, how do you think about customer success versus sales as an organization, and how do you organize that internally, right? So, so repricing can have lots of impacts, not just on your end user, but also on the internal dynamics of your org.

[12:59](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=779s) Um, fi- finally, the thing I'll say is, uh, as you're iterating, you, you really want to, as much as possible, be able to simulate the impact that this pricing change is gonna have on your users.

[13:09](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=789s) In particular, I think, uh, you wanna be able to simulate, uh, different usage patterns. So if you think about your users in cohorts, um, some might be using it for very different use cases.

[13:20](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=800s) What is that revenue mix gonna look like? What is your top, uh, revenue-generating customer going to... How, how are they gonna change with a new pricing model? You, you wanna be able to do that in a, in a data-informed way, especially if you're running at a, at a reasonable scale.

[13:34](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=814s) Here are some predictions I have as, as we're wrapping up for AI agent pricing. I think, um, you know, the, the price wars will continue. We're seeing agents get cheaper and cheaper.

[13:43](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=823s) Um, and, and, you know, I think you're, you-- we should expect, uh, people to try to give effectively unlimited plans. I think there'll still be caps, I think there'll still be guardrails, but I think, um, w- we, we will move to a world where, uh, you know, pe- people just want these agents embedded in every workflow, so

[14:01](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=841s) try to, try to converge to effectively unlimited. Um, [lip smack] I, I think outcome-based pricing is going to get more real. We are seeing some companies, uh, do that already. But the, the, the hard part about outcome-based pricing is defining what the actual SLA is.

[14:15](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=855s) Like, what, what is the outcome, and how do you measure it, and does it make sense to the end user? Um, I, I think, like, we're gonna see a lot of clear definition of success language in contracts, and, like, people really trying to get, uh, precise about what the outcome is that, that they're charging for.

[14:32](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=872s) And then, um, I, I think there's gonna be a lot more real-time visibility, spend management, uh, balance alerts. I, I think the kind of workflow around pricing and monetization is gonna get, uh, more and more sophisticated.

[14:45](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=885s) You, you think about, like, you know, obviously in, in database queries, you have, like, query planners. I think it's quite possible that, you know, you, you prompt an agent, it tells you, like, "Here's my estimate for how much, uh, how many credits I'm gonna use," or, "Hey, here's three options for, um, how I might execute, how I

[15:02](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=902s) might plan this workload. W- one's gonna take 10,000 credits, the other's gonna take, you know, 4,000 credits." I think that customer experience of, like, being able to control the workflow around the spend is likely to get much, much more sophisticated, um, this year.

[15:17](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=917s) So, uh, just, just final note, um, we have a whole e-book on how to price your AI agent that goes into much more tactical advice, talks about specific, you know, threshold billing, fraud, prepaid pricing models.

[15:28](https://www.youtube.com/watch?v=1C3sZbaxOmw&t=928s) Um, if you wanna check that out, that is at this link. Thank you. [audience applauding] [upbeat music]
