# We're the bottleneck, but we don't have to be — Ido Salomon, AgentCraft Ido Salomon, AgentCraft AI Engineer World's Fair 2026 [Talk page](https://ai.engineer/talks/YIVkERhy8xo-i-turned-coding-agents-into-strategy-game) · [Video](https://www.youtube.com/watch?v=YIVkERhy8xo) ## We're the bottleneck, but we don't have to be AgentCraft borrows the controls of strategy games to make agent work visible, move attention to blocked tasks, and share supervision. Its next experiment asks how much of that machinery a new user should need to see. ### An army of agents still needs someone to direct it [0:21](https://www.youtube.com/watch?v=YIVkERhy8xo&t=21s) Coding agents can produce anything from cat memes to B2B SaaS applications. Why, then, doesn’t everyone have an army of them? Ido Salomon, creator of AgentCraft, starts with the gap between an agent’s capability and a person’s ability to use many agents at once. The imagined setup is easy: get enough monitors, start 25 agents, and let them work. The actual work begins after they start. ![Source frame: An army of agents still needs someone to direct it](https://ai.engineer/api/talk-essay-media/e5d7af76b78f24deff158d8802f9f45c7d354beea3515eaaa1320b4a10bb60be) Source frame: An army of agents still needs someone to direct it Each agent needs steering, direction, and review. Adding agents adds demands on the same person’s attention, until supervision becomes exhausting. Salomon’s diagnosis is that the human becomes the bottleneck. More work can run concurrently than one person can comfortably understand and guide. Strategy games offer a familiar way to handle this problem. In Warcraft, or even The Sims, multiple units act while the player watches, intervenes, and decides what should happen next. AgentCraft takes that experience into work: a game-inspired orchestrator that represents agents spatially and gives their supervisor controls for understanding and directing them. The useful skill being borrowed is managing several active units without inspecting every action continuously. ### Turn agent activity into a map you can act on [2:41](https://www.youtube.com/watch?v=YIVkERhy8xo&t=161s) An AgentCraft unit represents an actual agent. The application can detect agents already running on the device or start new ones. Selecting a unit opens a side panel where the user can prompt it, including by voice. Buildings represent other work functions: managing plugins and skills, opening an integrated terminal, and using Git. The game vocabulary organizes ordinary tools into one space. ![Source frame: Turn agent activity into a map you can act on](https://ai.engineer/api/talk-essay-media/6c42c9fca60555f6bcb7d88808a3171ae6fd666851d9c3bce08bc715ae680cea) Source frame: Turn agent activity into a map you can act on Visibility has several distinct jobs: - **Agent status:** The side panel shows each agent’s task, its last action, and what it is doing now. That gives the user a quick way to decide who needs attention. - **Location of work:** The file system can be projected onto the map. Files appear as runes, and agent activity is associated with the directories or files being worked on. Inspecting a file reveals what an agent did and when. - **Activity overview:** Those associations support heat maps, letting the user see where work is happening across the project. A map helps only if the user can respond to what it reveals. AgentCraft borrows another game control: pressing the space bar jumps to the next nearby item needing attention. The supervisor can move between questions and plan approvals rather than repeatedly searching through every agent. Status tells the user what is happening; the shortcut turns that information into an intervention. ### Quests start work; autonomy moves the burden to review [5:34](https://www.youtube.com/watch?v=YIVkERhy8xo&t=334s) Seeing everything does not mean being able to hold everything in your head. Following agents at the level of files and tool calls remains tiring, and a large project creates another demand: deciding what to do next. AgentCraft lets an agent inspect the codebase and suggest tasks. Those tasks become quests that the user can accept with a button, after which an agent does the work. ![Source frame: Quests start work; autonomy moves the burden to review](https://ai.engineer/api/talk-essay-media/b29b482f8a0ca7a6e169bc5a2a3513226e9cd218df22ae8a9d80b1c524ca6a2a) Source frame: Quests start work; autonomy moves the burden to review Task suggestions relieve the user of inventing every next step, but accepting 20 quests can still leave 20 jobs to babysit. The next layer delegates coordination itself. A user supplies a general goal; an orchestrator breaks it down and runs the work. In the demonstrated setup, execution happens in a local container, which Salomon describes as keeping the work isolated while the agent handles supervision. Loops extend that delegation to ongoing sources of work. The examples are scanning Twitter for things to build or scanning GitHub, then creating features in the background. This reduces how much the user must specify up front. It also creates a downstream problem: even five completed changes arriving together can be difficult to review. The review kit packages the result for that decision. It offers the changes and changed files for file-by-file inspection, alongside images or videos showing what changed. It also supports comparing implementations produced by several parallel instances and choosing the preferred result. Visual evidence makes the visible consequence easier to inspect; the file view remains available for examining the underlying changes. Where does human attention go as more work becomes autonomous? The flow below shows two ways work enters execution: a general goal decomposed by an orchestrator, and background loops that find work. Both eventually produce changes for review. Delegating execution reduces continuous supervision, while making the presentation of completed work more important. From delegated work to a review decision Orchestrators and loops reduce ongoing direction, but their results still converge on a review step. - General goal — The user describes what they want. - Orchestrator — Breaks down and runs work in the demonstrated local container. - Background loops — Scan Twitter or GitHub and create features. - Review kit — Changed files plus images or videos of the result. - Human review — Inspect changes or compare parallel implementations. - General goal → Orchestrator (Delegate decomposition) - Orchestrator → Review kit (Completed changes) - Background loops → Review kit (Background results) - Review kit → Human review (Inspect and compare) ### Let design and implementation continue together [8:11](https://www.youtube.com/watch?v=YIVkERhy8xo&t=491s) Visibility and autonomy still leave one person responsible for the results. That person may also be the wrong reviewer for part of the work. AgentCraft’s next step is collaboration: locally hosted shared rooms, with tunnels allowing other people to join. ![Source frame: Let design and implementation continue together](https://ai.engineer/api/talk-essay-media/5fd662a78a328648a6981bb350cbeec66e923b991b3e86328ca0c3b774c61d6f) Source frame: Let design and implementation continue together The concrete example is a product designer joining the room—in this case, Salomon’s wife. After he approves her entry, she works on a design. Her work becomes visible on his map, so he can follow up on the design and implement it while she continues working. The observable change is from one person supervising alone to two people advancing different parts of the same project concurrently. The shared workspace supports handing off or forking work. Git remains usable, while coordination also happens inside the room. Chat connects people and agents, and a notice board records what each participant is doing. In the design example, the engineer can announce that he has taken on the design task, giving the others information about the follow-up rather than leaving them to infer it from changes. How can implementation begin without making the designer stop? The diagram follows the example’s two paths: the designer continues designing while the engineer follows up on the visible work. The room’s activity information helps participants coordinate those paths. Access also extends to mobile devices and Telegram, so participation need not stay tied to the map on a desktop. A design becomes work the engineer can follow up on Visibility enables a handoff while the designer continues; room updates tell others what work has been taken on. - Designer joins — Entry is approved, then design work begins. - Design visible on the map — The engineer sees the designer’s work. - Designer continues — Design work proceeds alongside implementation. - Engineer implements — Follows up on the design. - Room activity update — The engineer announces taking on the design task. - Designer joins → Design visible on the map (Work becomes visible) - Design visible on the map → Engineer implements (Follow up on design) - Designer joins → Designer continues (Keep working) - Engineer implements → Room activity update (Announce the task) ### A simpler game for people who do not want file-level controls [10:33](https://www.youtube.com/watch?v=YIVkERhy8xo&t=633s) The combination of visibility, autonomy, and collaboration raises the ceiling for power users. The more surprising feedback concerns people who had not used agents before: parents reporting that their children enjoyed orchestrating agents, and a former StarCraft player finding a productive use for familiar skills. These are Salomon’s reported anecdotes; his estimate that roughly 90% of people need a lower entry point is an approximation, rather than a measured adoption result. ![Source frame: A simpler game for people who do not want file-level controls](https://ai.engineer/api/talk-essay-media/211ea4789bd7c75d4959cff0bbf7c64726a691082974c2afd5fa8465104d3797) Source frame: A simpler game for people who do not want file-level controls That feedback leads to a different interface experiment, tentatively called Loopers. Warcraft-style complexity can itself be daunting, especially for someone uninterested in which files an agent is editing. The proposed experience moves toward the level of a mobile game: projects users return to, autonomous loops that keep work going, and a basic interaction of prompting and receiving a result. The tradeoff is how much execution detail the interface asks the user to manage. AgentCraft brings file activity forward; the simpler concept lets users focus on results and inspect the process when they want to. It relies more heavily on agents acting autonomously, while trying to make the experience inviting enough that people return to their projects. A game interface will not feel easy to everyone. Something inviting to one person can be difficult for another. Salomon therefore treats this as an early experiment in customization: finding what a particular person needs to understand the tools and want to use them. Loopers is a very early draft, and its tentative name reflects that unfinished product direction. ### Choose the level of supervision you want [13:24](https://www.youtube.com/watch?v=YIVkERhy8xo&t=804s) The closing invitation separates the two products’ stages. AgentCraft is already available to use. The simpler concept remains experimental, with Salomon seeking interest and feedback about the level of accessibility or capability people want. The next design decision depends on what users need to see, control, and return to—not on assuming that one game metaphor suits everyone. The practical invitation is to take familiar skills from elsewhere in life and apply them to agent work. Strategy games supply one such vocabulary: watch units, notice what needs attention, delegate work, and coordinate with others. AgentCraft turns those actions into controls for coding agents; the simpler experiment asks which controls a person needs before the experience becomes useful rather than exhausting. ## Key takeaways - Multiple agents create a human coordination problem: each adds steering, direction, and review demands. - AgentCraft combines status, spatial file activity, heat maps, and an attention shortcut so users can understand work and intervene. - Quests and orchestrators reduce planning and supervision work; the review kit makes their completed changes easier to inspect and compare. - Shared rooms let specialists follow up on each other’s work while continuing in parallel, with activity updates connecting people and agents. - The experimental Loopers concept reduces required file-level detail, but accessibility depends on the person; customization remains part of the exploration. ## Slides ### We’re the bottleneck, but we don’t have to be A human sits at an approval gate along a conveyor of glowing blocks, surrounded by fantasy workers and a connected world map. Salomon introduces the question of why people become bottlenecks. [0:22](https://www.youtube.com/watch?v=YIVkERhy8xo&t=22s) [Fantasy illustration of a human overseeing an automated workflow and a world map.](https://assets.ai.engineer/sites/aiecode2025/sha256/3b1256b2d0746602164ed421fc2661611f00e496f14f457992eecaf9b069bff1.webp) We’re the bottleneck, but we don’t have to be Ido Salomon AI Engineer World’s Fair Interface text: CC ### Ido Salomon Ido Salomon introduces his work on AgentCraft and MCP projects, then asks how people can steer and review multiple agents without burning out. [0:36](https://www.youtube.com/watch?v=YIVkERhy8xo&t=36s) [Portrait of Ido Salomon with his AgentCraft, MCP-UI and MCP Apps roles.](https://assets.ai.engineer/sites/aiecode2025/sha256/a8a339d06b28dfa4f7292ac28c27530b3617c3378d8e57489df6aa5baade7669.webp) Ido Salomon Creator @ AgentCraft Creator @ MCP-UI Co-Creator and Maintainer @ MCP Apps X @idosal1 idosal in ido-salomon Interface text: CC ### We are the bottleneck Salomon says steering, directing and reviewing each agent becomes exhausting at scale, making people the bottleneck. [1:37](https://www.youtube.com/watch?v=YIVkERhy8xo&t=97s) [The statement “We are the bottleneck” on a dark framed background.](https://assets.ai.engineer/sites/aiecode2025/sha256/ecb0b107cb90a7308b1cf9953b2531a60cf32d6c64e28632ce6fc03ae5a2c8b9.webp) We are the bottleneck Interface text: CC ### These skills were with us all along Salomon compares supervising agents to tracking and supervising units in Warcraft or Sims, suggesting that familiar game skills can help people work with agents without burning out. [1:42](https://www.youtube.com/watch?v=YIVkERhy8xo&t=102s) [The statement “These skills were with us all along” on a dark framed background.](https://assets.ai.engineer/sites/aiecode2025/sha256/33fe4c98766920328b111966a80526e954b53d6828e70d78a9fb78f882ee12fc.webp) These skills were with us all along Interface text: CC ### AgentCraft: Putting the Orc in Agent Orchestration Salomon compares supervising agents to managing game units and introduces AgentCraft as a game-inspired orchestrator. The title slide shows orcs operating computers around a world map. [2:16](https://www.youtube.com/watch?v=YIVkERhy8xo&t=136s) [Fantasy illustration of orcs operating computers around a world map.](https://assets.ai.engineer/sites/aiecode2025/sha256/d8cb03b3f57349b14fffd28b7625090f8eea797fba2ef8545867a7c4fa7140b2.webp) AGENTCRAFT Putting the Orc in Agent Orchestration ### AgentCraft overview AgentCraft displays a battlefield with agents and buildings alongside panels for agent tasks, recent activity, and alliance controls. [2:40](https://www.youtube.com/watch?v=YIVkERhy8xo&t=160s) [A game-like battlefield with agent activity and alliance panels.](https://assets.ai.engineer/sites/aiecode2025/sha256/8a083e057f48094515f098e5510a9191dc720caea95fc05c29f0f1350397d651.webp) tests/ tsc-runner/ supabase/ [illegible]-build/ shared/ server/ [illegible]t-report/ [illegible]ywright-build/ lib/ frontend/ .claude/ .codex/ .cursor/ .github/ .playwright-mcp/ Interface text: AGENTCRAFT 56 3 / 0 44.5M $184.7 MY AGENTS All 4 Awaiting Working 3 Idle 1 DISMISS ALL Mag'tuk Add beads.dev task management int... Running grep [illegible] Mag'grim Submit campaign to new section a... Reading planDecomposer.ts Kilro'ash Add channels feature to README Running grep [illegible] Rek'osh Idle 28m ACTIVITY (9) Clear Quest Started 55s Quest Started 1m Quest Started 1m ALLIANCE PANEL Remote Access Open › Share your battlefield via link Channels Open › 1 bot configured for pairings Hall Open › Multi-machine agent federation ACTIVE LOOPS ### Agent character An orc represents a coding agent on the battlefield, while a side panel tracks agents’ tasks and activity. Salomon says agents can be detected on the device or spawned from AgentCraft. [2:50](https://www.youtube.com/watch?v=YIVkERhy8xo&t=170s) [An orc agent stands on the battlefield beside activity panels.](https://assets.ai.engineer/sites/aiecode2025/sha256/fd5669656e412300a01761081e225a964d049ef70304f302546a4af6283fd917.webp) MY AGENTS All 4 Awaiting Working 3 Idle 1 DISMISS ALL Mag'tuk Add beads.dev task management int... Reading missionManager.ts Mag'grim Submit campaign to new section a... Reading planMemoryManager.ts Kilro'ash Add channels feature to README Rek'osh Idle 29m ACTIVITY (9) Clear Quest Started 1m Quest Started 1m Quest Started 1m Hero Recruited 2m ALLIANCE PANEL Remote Access Open › Share your battlefield via link Channels Open › 1 bot configured for pairings Hall Open › Multi-machine agent federation ACTIVE LOOPS Interface text: AGENTCRAFT 56 3/10 44.5M $184.7 ### Great Hall session roster Selecting the Great Hall brings up a session roster beneath the battlefield, while the agent panel shows active tasks. The speaker says agents can be detected on the device or spawned from AgentCraft. [3:04](https://www.youtube.com/watch?v=YIVkERhy8xo&t=184s) [The Great Hall is selected on the battlefield, with a session roster below.](https://assets.ai.engineer/sites/aiecode2025/sha256/0bf4e0f2d93fbdb488471701c1a88fa6a884d7229aaf043ee2524f05be9a102a.webp) SESSION ROSTER Right-side panel shift fix summary main OFF Post-completion task planning: next steps after right-panel fix main OFF Right-side panel shift fix JSON summary OFF 50 sessions VIEW ALL SESSIONS GREAT HALL Interface text: taskDispatcher.ts verificationManager.ts MY AGENTS All 4 Awaiting Working 3 Idle 1 DISMISS ALL Mag'tuk — Add beads.dev task management int... Running grep -n "app\.get\|app\.post\|app\.patch... Mag'grim — Submit campaign to new section a... Reading taskDispatcher.ts Kilro'ash — Add channels feature to README Writing staged-plotting-wind.md Rek'osh — Idle 29m ACTIVITY (9) Clear Quest Started — 1m Quest Started — 1m Quest Started — 1m ALLIANCE PANEL Remote Access — Open › Share your battlefield via link Channels — Open › 1 bot configured for pairings Hall — Open › Multi-machine agent federation ACTIVE LOOPS 56 3 1 0 44.5M $184.7 LAUNCH A CURSOR AGENT IN CURREN[illegible] ### Prompting an agent The presenter prompts Durg’dur through the side panel, which also has voice controls. After the request to add Copilot CLI support, the agent switches from idle to working and begins planning. [3:15](https://www.youtube.com/watch?v=YIVkERhy8xo&t=195s) [A selected agent appears beside an empty chat panel with message and voice controls.](https://assets.ai.engineer/sites/aiecode2025/sha256/eff4e79c2f5a1478c9d3186158b420add8373ef4408e83bed09fb02fa07a9ea6.webp) Durg'dur DEFAULT Idle No active mission Start a conversation Message... DEFAULT PLAN MODE Interface text: MY AGENTS All 5 Awaiting 1 Working 2 Idle 2 DISMISS ALL Kilro'ash — Add channels feature to README Awaiting plan approval Mag'tuk — Add beads.dev task management int... Running grep -n "app\.get\|app\.post\|app\.patch... Mag'grim — Submit campaign to new section a... Reading CampaignPanel.tsx Durg'dur — Idle 3s Rek'osh — Idle 29m ACTIVITY (10) Clear Hero Recruited — 3s Quest Started — 1m ALLIANCE PANEL Remote Access — Open › Share your battlefield via link Channels — Open › 1 bot configured for pairings Hall — Open › Multi-machine agent federation ACTIVE LOOPS CHAT GIT FILES IDLE AWAITING ORDERS MODEL — No messages yet Send a message... DURG'DUR 56 44.5M $184.7 [3:26](https://www.youtube.com/watch?v=YIVkERhy8xo&t=206s) [Game-style agent dashboard with a submitted task and a planning indicator in the chat panel.](https://assets.ai.engineer/sites/aiecode2025/sha256/e76b5f96ae96931ded7e654087e35f8591f3ecded3435c8c49a539981b2753a7.webp) Durg... DEFAULT [illegible] Working 4s add copilot cli support 0% ctx YOU add copilot cli support Planning the implementation... Type to queue message... DEFAULT PLAN MODE Interface text: MY AGENTS All 5 Awaiting 1 Working 3 Idle 1 DISMISS ALL Kilro'ash — Add channels feature to README Add Channels feature to README Durg'dur — add copilot cli support Mag'tuk — Add beads.dev task management int... Mag'grim — Submit campaign to new section a... Reading CampaignPanel.tsx Rek'osh — Idle 29m ACTIVITY (11) Clear Quest Started — 4s Hero Recruited — 15s Quest Started — 1m ALLIANCE PANEL Remote Access — Open › Share your battlefield via link Channels — Open › 1 bot configured for pairings Hall — Open › Multi-machine agent federation ACTIVE LOOPS CHAT GIT FILES add copilot cli support Mission: add copilo... MODEL Opus YOU add copilot cli support Agent is thinking... DURG'DUR 56 44.5M $184.7 ### Extensions view Map buildings represent functions: the selected War Factory shows an Extensions view with 15 plugins and no configured MCP servers. [3:34](https://www.youtube.com/watch?v=YIVkERhy8xo&t=214s) [Game-style workspace with an extensions panel showing 15 plugins and no MCP servers.](https://assets.ai.engineer/sites/aiecode2025/sha256/1f7fe4ea57a73336ff293595a3a046ffe0978f6c144997200769416d02119ab4.webp) EXTENSIONS 15 Plugins frontend-design, typescript-lsp 0 MCP Servers None configured 15 plugins, 0 servers Interface text: MY AGENTS All 5 Awaiting 1 Working 3 Idle 1 DISMISS ALL Kilro'ash — Add channels feature to README Add Channels feature to README Durg'dur — Add Copilot CLI platform support Mag'tuk — Add beads.dev task management int... Mag'grim — Submit campaign to new section a... Reading CampaignPanel.tsx Rek'osh — Idle 29m ACTIVITY (11) Clear Quest Started — 7s Hero Recruited — 18s Quest Started — 1m ALLIANCE PANEL Remote Access — Open › Share your battlefield via link Channels — Open › 1 bot configured for pairings Hall — Open › Multi-machine agent federation ACTIVE LOOPS WAR FACTORY lib/ .claude .codex/ .cursor/ .github/ AGENTCRAFT 56 44.5M $184.7 ### Agent roster AgentCraft places agent portraits beneath the map and tracks their tasks in a side panel. The roster shows three working agents, one awaiting input, and one idle; buildings represent workspace functions. [3:47](https://www.youtube.com/watch?v=YIVkERhy8xo&t=227s) [Game-style workspace with a grid of agent portraits below the map.](https://assets.ai.engineer/sites/aiecode2025/sha256/3ac92c53c599e9001965ad5b6487a388e4d40448ab60409f5a612b9e015ae7ef.webp) MY AGENTS All 5 Awaiting 1 Working 3 Idle 1 DISMISS ALL Kilro'ash — Add channels feature to README Add Channels feature to README Durg'dur — Add Copilot CLI platform support Mag'tuk — Add beads.dev task management int... Mag'grim — Submit campaign to new section a... Rek'osh — Idle 29m ACTIVITY (11) Clear Quest Started — 22s Hero Recruited — 33s Quest Started — 1m Quest Started — 2m ALLIANCE PANEL Remote Access — Open › Share your battlefield via link Channels — Open › 1 bot configured for pairings Hall — Open › Multi-machine agent federation ACTIVE LOOPS Map labels: lib/ .claude .codex/ .cursor/ .github/ Interface text: 56 3/1/1 44.5M $184.7 AGENTCRAFT ### Agent status The agent panel shows who is working, idle or awaiting input. Hovering over Kilro’ash reveals its mission, context usage and recent activity, helping the user identify who needs attention. [3:56](https://www.youtube.com/watch?v=YIVkERhy8xo&t=236s) [An agent tooltip shows an awaiting-input status, mission, context usage and recent activity.](https://assets.ai.engineer/sites/aiecode2025/sha256/34b927d199f82d5f399526f2b8ed60933bd7b0574b91bee778cbb6d339bb9ce4.webp) MY AGENTS All 5 Awaiting 1 Working 3 Idle 1 DISMISS ALL Kilro'ash — Add channels feature to README Add Channels fe[illegible] Durg'dur — [illegible] Mag'tuk — Add be[illegible] Mag'grim — Subm[illegible] Rek'osh — Idle 29m Kilro'ash Status: Awaiting Input Mission: Add channels feature to README Context: 17% used Active: 3m ago ACTIVITY (11) Clear Quest Started — 25s Map labels: supabase/ srv-build/ shared/ [illegible]erver/ [illegible]port/ Interface text: 56 3/1/1 ### Agents on the map AgentCraft places agents among project directories while a panel tracks their status. Salomon describes using orchestration data to create heat maps and see what is happening. [4:58](https://www.youtube.com/watch?v=YIVkERhy8xo&t=298s) [A game-like map displays agents, project areas and a list with two agents awaiting input.](https://assets.ai.engineer/sites/aiecode2025/sha256/5b2f97f080fcfe7930b0908d3a64d53ba98b9763ea015c7810b84563eca69cf6.webp) MY AGENTS All 5 Awaiting 2 Working 2 Idle 1 DISMISS ALL Mag'grim Submit campaign to new section a... What do you mean by Kilro'ash Add channels feature to README Add Channels feature to README Durg'dur Add Copilot CLI platform support Running git log --oneline --all | grep -i "goose... Mag'tuk Add beads.dev task management int... Rek'osh Idle 30m ACTIVITY (11) Clear Quest Started 1m Hero Recruited 1m ALLIANCE PANEL Remote Access Open › Share your battlefield via link Channels Open › 1 bot configured for pairings Hall Open › Multi-machine agent federation ACTIVE LOOPS + Map labels: tests/ tsc-runner/ supabase/ srv-build/ shared/ server/ scripts/ plugin/ playwright-report/ playwright-build/ lib/ frontend/ .claude/ .codex/ .cursor/ .github/ .playwright-mcp/ Interface text: AGENTCRAFT 56 2 1/2 44.5M $184.7 ### Reviewing agent requests Salomon describes using the spacebar to jump to the nearest item needing attention. The demo shows a clarification question and a plan with approval choices. [5:08](https://www.youtube.com/watch?v=YIVkERhy8xo&t=308s) [An agent chat panel asks for clarification about a new section and offers two response choices.](https://assets.ai.engineer/sites/aiecode2025/sha256/5ee241fb5ffb0e65fbd8f069b0041f305e1f5261a36834aad9a2c54612f04574.webp) Submit campaign to new section as item current state of the modified CampaignPanel to understand what changes are already in progress, then clarify with you. Let me read more of the file to see the full section structure. I need to clarify the request. Currently, when a feature is submitted, it shows a generic “Analyzing spec...” banner (line 285-289) but doesn’t display the generating plans as individual items in a section like the other plan statuses do (Awaiting Review, Active, Completed, etc.). WHAT DO YOU MEAN BY ‘NEW SECTION’? CURRENTLY, SUBMITTED PLANS THAT ARE GENERATING SHOW AS A BANNER, NOT AS ITEMS IN A SECTION. SHOULD THEY BECOME ITEMS IN A NEW ‘GENERATING’ SECTION? OR DO YOU MEAN SOMETHING ELSE ENTIRELY? 1 Generating section 2 Different section Type your answer... Interface text: MY AGENTS All 5 | Awaiting 2 | Working 2 | Idle 1 DISMISS ALL Mag'grim — Submit campaign to new section a... What do you mean by Kilro'ash — Add channels feature to README Add Channels feature to README Durg'dur — Add Copilot CLI platform support Running grep -A2 [illegible]... Mag'tuk — Add beads.dev task management int... Rek'osh — Idle 30m ACTIVITY (11) Clear Quest Started — 1m Hero Recruited — 1m ALLIANCE PANEL Remote Access — Open › Share your battlefield via link Channels — Open › 1 bot configured for pairings Hall — Open › Multi-machine agent federation ACTIVE LOOPS scripts/ plugin/ playwright-report/ playwright-build/ lib/ frontend/ fe-build/ docs/ cloudflare/ CHAT | GIT | FILES Mag'g... | DEFAULT (O... | Awaiting DEFAULT | PLAN MODE MAG'GRIM Mission: Submit cam... MODEL: Opus SCOUTS 3 [5:12](https://www.youtube.com/watch?v=YIVkERhy8xo&t=312s) [AgentCraft shows an agent’s plan and approval options beside the world map.](https://assets.ai.engineer/sites/aiecode2025/sha256/9bea07d439b47f6c114402711611d7800dea2fb971d6ca99d2bf2a8c96b4a59b.webp) Add channels feature to README PLAN Review the plan and choose how to proceed. Plan: Add Channels Feature to README Context The channels feature lets users communicate with their AgentCraft heroes directly through messaging apps (Telegram, Discord, Slack, WhatsApp). It supports both hero- Yes, clear context Yes, keep context Yes, manual approval Decline Type feedback on the plan... Interface text: MY AGENTS All 5 | Awaiting 2 | Working 2 | Idle 1 DISMISS ALL Mag'grim — Submit campaign to new section a... What do you mean by Kilro'ash — Add channels feature to README Add Channels feature to README Durg'dur — Add Copilot CLI platform support Running grep [illegible]... Mag'tuk — Add beads.dev task management int... Rek'osh — Idle 30m ACTIVITY (11) Clear Quest Started — 1m Hero Recruited — 1m ALLIANCE PANEL Remote Access — Open › Share your battlefield via link Channels — Open › 1 bot configured for pairings Hall — Open › Multi-machine agent federation ACTIVE LOOPS tasks/ test-runner/ tests/ tsc-runner/ .agentcraft/ .agents/ .campaign/ .claude/ .codex/ .cursor/ .github/ .playwright-mcp/ CHAT | GIT | FILES Kilro... | DEFAULT (O... | Awaiting 5 sections Show conversation DEFAULT | PLAN MODE KILRO'ASH Mission: Add channe... MODEL: Opus SCOUTS 1 (line ~78, after Alliance Hall) ## Verification - Read the final README to confirm formatting and placement look correct Send a message... ### AgentCraft overview Salomon describes using the spacebar to jump to the nearest item needing attention, then answer questions or approve plans. The map and agent list show agents and their current tasks. [5:17](https://www.youtube.com/watch?v=YIVkERhy8xo&t=317s) [AgentCraft displays agent activity, a world map, and a roster of characters.](https://assets.ai.engineer/sites/aiecode2025/sha256/82ddcf5e188b11f5ed89b84d4ac94d0d92ad84d4378e325191122019f495303b.webp) tasks/ test-runner/ tests/ tsc-runner/ .agentcraft/ .agents/ .campaign/ .claude/ .codex/ .cursor/ .github/ .playwright-mcp/ Interface text: AGENTCRAFT 56 2 1/2 44.5M $184.7 MY AGENTS All 5 Awaiting 2 Working 2 Idle 1 DISMISS ALL Mag'grim Submit campaign to new section a... What do you mean by Kilro'ash Add channels feature to README Add Channels feature to README Durg'dur Add Copilot CLI platform support Reading index.ts Mag'tuk Add beads.dev task management int... Running gh api repos/steveyegge/beads/readme -j... Rek'osh Idle 30m ACTIVITY (11) Clear Quest Started 1m ALLIANCE PANEL Remote Access Open › Share your battlefield via link Channels Open › 1 bot configured for pairings Hall Open › Multi-machine agent federation ACTIVE LOOPS ### Visibility A map highlights one castle while a question mark marks another location. Salomon says visibility helps, but tracking each agent’s files and tools is exhausting. [5:32](https://www.youtube.com/watch?v=YIVkERhy8xo&t=332s) [A fantasy map with a highlighted castle, a question mark, and the word Visibility.](https://assets.ai.engineer/sites/aiecode2025/sha256/c927b2f069e2153491820552f50b3fae85d384560df53748c0a72a7447bb83b9.webp) Visibility ? Interface text: CC ### Too much to hold in our heads Even with visibility into agents, tracking a large project and deciding what to do next can overwhelm one person. Salomon suggests asking agents to find tasks and turn them into quests. [5:55](https://www.youtube.com/watch?v=YIVkERhy8xo&t=355s) [A slide about cognitive load with a Shrek quest meme.](https://assets.ai.engineer/sites/aiecode2025/sha256/0d9fb4c1aadfc8b340862e488eb77bd51f7f672b1b0ffc8f1618baeee7892dd2.webp) Too much to hold in our heads ARE YOU UP FOR A LITTLE QUEST, DONKEY? Interface text: CC ### Accepting quests Salomon describes agents scanning the codebase for tasks, then taking on quests the user accepts. Supervising many simultaneous quests, he says, becomes exhausting. [6:17](https://www.youtube.com/watch?v=YIVkERhy8xo&t=377s) [A quest list shows suggested validation and rate-limiting tasks with Accept Quest buttons.](https://assets.ai.engineer/sites/aiecode2025/sha256/bc9768c91493097812ee53ac83ae8df5ca2f249df5a403e4d91ae726d3501c32.webp) Fortification 2 Proving Grounds 1 Restoration 51 Purification 2 Fortification (2) Environment & API Payload Validation with ... Elite The server has no schema validation for environment variables or incoming API pa... Fortification Accept Quest WebSocket Event Rate Limiting Elite HTTP routes have express-rate-limit but WebSocket messages have no rate limiting... Fortification Accept Quest ### Submit a campaign feature The demo opens a form for describing a feature. Salomon says an orchestrator can break down the request and run it in a local container. [6:47](https://www.youtube.com/watch?v=YIVkERhy8xo&t=407s) [The Campaign panel shows a submit feature button, completed count, and history.](https://assets.ai.engineer/sites/aiecode2025/sha256/d4bd30e4a33e64b0095a162e8a0b1fe8a3d9c167d6218849b463905bcbb5d777.webp) QUESTS(56) CAMPAIGN TRACES ANALYTICS MEMORY + SUBMIT FEATURE Done: 9 HISTORY (3) Add approval and kick for remote access (like war rooms) MERGED Review Fix quick action buttons overflow when side panel narrows CANCELLED x Fix quick action buttons overflow when side panel narrows CANCELLED x COMPLETED (9) Interface text: MY AGENTS All Awaiting Working Idle ACTIVITY (11) Quest Started ALLIANCE PANEL Remote Access Channels Hall ACTIVE LOOPS [6:50](https://www.youtube.com/watch?v=YIVkERhy8xo&t=410s) [Campaign panel with an empty feature description form.](https://assets.ai.engineer/sites/aiecode2025/sha256/3ce00dcf2376cf9e40288d28a432f2df05ffe9c7fc3d53f24c18225243cab85b.webp) QUESTS(56) CAMPAIGN TRACES ANALYTICS MEMORY + SUBMIT FEATURE Describe the feature you want to build... CANCEL SUBMIT Done: 9 HISTORY (3) Add approval and kick for remote access (like war rooms) MERGED Review ### Campaign planning The Campaign panel shows an agent analyzing a spec. Salomon says an orchestrator breaks down a broad request and runs the work in a local container. [6:57](https://www.youtube.com/watch?v=YIVkERhy8xo&t=417s) [Campaign panel showing an agent analyzing a specification and a history of features.](https://assets.ai.engineer/sites/aiecode2025/sha256/33ab6962b2db5b441f9551f5d139c574ee614fa4bccda884a45b4990d8b27bf7.webp) QUESTS (56) CAMPAIGN TRACES ANALYTICS MEMORY + SUBMIT FEATURE Done: 9 ANALYZING SPEC... AN AGENT IS PLANNING YOUR CAMPAIGN. HISTORY (3) Add approval and kick for remote access (like war rooms) MERGED Review Fix quick action buttons overflow when side panel narrows CANCELLED × Fix quick action buttons overflow when side panel narrows CANCELLED × COMPLETED (9) Interface text: MY AGENTS All 7 Awaiting 2 Working 3 Idle 2 DECOMPOSING... Zorn'mar Orchestrator Mag'grim Submit campaign to new section a... What do you mean by Kilro'ash Add channels feature to README Add Channels feature to README Drak'dan Working... Durg'dur Add Copilot CLI platform support Mag'tuk Add beads.dev task management int... Rek'osh Idle 31m ALLIANCE PANEL Remote Access Open › Share your battlefield via link Channels Open › 1 bot configured for pairings Hall Open › Multi-machine agent federation ACTIVE LOOPS ## Full transcript [0:21](https://www.youtube.com/watch?v=YIVkERhy8xo&t=21s) Okay. Hi everyone. Uh I'm Solomon. I am [0:24](https://www.youtube.com/watch?v=YIVkERhy8xo&t=24s) the creator of ATMCraft and I also [0:26](https://www.youtube.com/watch?v=YIVkERhy8xo&t=26s) created MCPUI which became MCP apps and [0:29](https://www.youtube.com/watch?v=YIVkERhy8xo&t=29s) MCP steering committee. I've done a [0:31](https://www.youtube.com/watch?v=YIVkERhy8xo&t=31s) bunch of other stuff. Uh but we're here [0:33](https://www.youtube.com/watch?v=YIVkERhy8xo&t=33s) to talk about why we're the bottleneck [0:36](https://www.youtube.com/watch?v=YIVkERhy8xo&t=36s) and why we don't have to be. [0:40](https://www.youtube.com/watch?v=YIVkERhy8xo&t=40s) So, uh as you all know and use every [0:43](https://www.youtube.com/watch?v=YIVkERhy8xo&t=43s) day, agents are amazing. There's so much [0:45](https://www.youtube.com/watch?v=YIVkERhy8xo&t=45s) they can do uh from generating cat memes [0:48](https://www.youtube.com/watch?v=YIVkERhy8xo&t=48s) to B2B SAS apps. So if they are so [0:52](https://www.youtube.com/watch?v=YIVkERhy8xo&t=52s) amazing, we have to ask why aren't we [0:54](https://www.youtube.com/watch?v=YIVkERhy8xo&t=54s) all unstoppable? Why doesn't everyone [0:56](https://www.youtube.com/watch?v=YIVkERhy8xo&t=56s) have an army of agents that just does [0:59](https://www.youtube.com/watch?v=YIVkERhy8xo&t=59s) whatever they want? So it should be [1:01](https://www.youtube.com/watch?v=YIVkERhy8xo&t=61s) relatively easy. I mean you just have a [1:04](https://www.youtube.com/watch?v=YIVkERhy8xo&t=64s) bunch of monitors and you spin up 25 [1:06](https://www.youtube.com/watch?v=YIVkERhy8xo&t=66s) cloud code agents and you're done. So [1:09](https://www.youtube.com/watch?v=YIVkERhy8xo&t=69s) the reality is that it's just not that [1:12](https://www.youtube.com/watch?v=YIVkERhy8xo&t=72s) simple to do something like that. you [1:14](https://www.youtube.com/watch?v=YIVkERhy8xo&t=74s) need to a really want to use agents and [1:18](https://www.youtube.com/watch?v=YIVkERhy8xo&t=78s) secondly once you actually spin them up [1:21](https://www.youtube.com/watch?v=YIVkERhy8xo&t=81s) each one of these agents requires you to [1:23](https://www.youtube.com/watch?v=YIVkERhy8xo&t=83s) steer them direct them review them and [1:26](https://www.youtube.com/watch?v=YIVkERhy8xo&t=86s) when it's at scale especially it's just [1:28](https://www.youtube.com/watch?v=YIVkERhy8xo&t=88s) exhausting we need to find the skills to [1:31](https://www.youtube.com/watch?v=YIVkERhy8xo&t=91s) allow us to do it without burning out so [1:34](https://www.youtube.com/watch?v=YIVkERhy8xo&t=94s) in reality we are the bottleneck [1:38](https://www.youtube.com/watch?v=YIVkERhy8xo&t=98s) and the the cool thing here is that [1:41](https://www.youtube.com/watch?v=YIVkERhy8xo&t=101s) while we are the bottleneck [1:43](https://www.youtube.com/watch?v=YIVkERhy8xo&t=103s) The skills that we need to do these [1:45](https://www.youtube.com/watch?v=YIVkERhy8xo&t=105s) things to uh find ourselves not burning [1:48](https://www.youtube.com/watch?v=YIVkERhy8xo&t=108s) out every time we use an agent are with [1:51](https://www.youtube.com/watch?v=YIVkERhy8xo&t=111s) us all along. Uh if you try to compare [1:55](https://www.youtube.com/watch?v=YIVkERhy8xo&t=115s) it to other stuff that you've been doing [1:57](https://www.youtube.com/watch?v=YIVkERhy8xo&t=117s) outside of programming, for example, [2:00](https://www.youtube.com/watch?v=YIVkERhy8xo&t=120s) playing video games, it's not really [2:01](https://www.youtube.com/watch?v=YIVkERhy8xo&t=121s) that different. Like if if you're [2:03](https://www.youtube.com/watch?v=YIVkERhy8xo&t=123s) playing Warcraft, for example, or Sims, [2:05](https://www.youtube.com/watch?v=YIVkERhy8xo&t=125s) and you have a bunch of different uh [2:07](https://www.youtube.com/watch?v=YIVkERhy8xo&t=127s) units running around, you need to [2:09](https://www.youtube.com/watch?v=YIVkERhy8xo&t=129s) understand what they're doing and [2:10](https://www.youtube.com/watch?v=YIVkERhy8xo&t=130s) supervise all of it. How different is [2:12](https://www.youtube.com/watch?v=YIVkERhy8xo&t=132s) that from from actually inspecting [2:14](https://www.youtube.com/watch?v=YIVkERhy8xo&t=134s) agents? So I took this concept uh too [2:18](https://www.youtube.com/watch?v=YIVkERhy8xo&t=138s) far and I created agent uh which is a [2:22](https://www.youtube.com/watch?v=YIVkERhy8xo&t=142s) game inspired orchestrator uh that tries [2:25](https://www.youtube.com/watch?v=YIVkERhy8xo&t=145s) to bring all that good stuff together [2:28](https://www.youtube.com/watch?v=YIVkERhy8xo&t=148s) into work. [2:30](https://www.youtube.com/watch?v=YIVkERhy8xo&t=150s) So let's take uh a quick look of what [2:32](https://www.youtube.com/watch?v=YIVkERhy8xo&t=152s) that actually looks like and try and [2:34](https://www.youtube.com/watch?v=YIVkERhy8xo&t=154s) understand the journey that we can go [2:36](https://www.youtube.com/watch?v=YIVkERhy8xo&t=156s) through to really raise the ceiling of [2:38](https://www.youtube.com/watch?v=YIVkERhy8xo&t=158s) how humans and agents can collaborate. [2:41](https://www.youtube.com/watch?v=YIVkERhy8xo&t=161s) So, uh, this is AgentCraft. Uh, there's [2:44](https://www.youtube.com/watch?v=YIVkERhy8xo&t=164s) really a lot to unpack here. Uh, so [2:46](https://www.youtube.com/watch?v=YIVkERhy8xo&t=166s) we'll just go through the basics really [2:48](https://www.youtube.com/watch?v=YIVkERhy8xo&t=168s) quick. Um, what you see here is an [2:51](https://www.youtube.com/watch?v=YIVkERhy8xo&t=171s) agent. Uh, it's an actual representation [2:55](https://www.youtube.com/watch?v=YIVkERhy8xo&t=175s) of a cloud code or openclaw, whatever [2:57](https://www.youtube.com/watch?v=YIVkERhy8xo&t=177s) agent that you can imagine. Uh, it can [3:00](https://www.youtube.com/watch?v=YIVkERhy8xo&t=180s) be detected on your device or you can [3:02](https://www.youtube.com/watch?v=YIVkERhy8xo&t=182s) actually spawn new ones directly from a [3:05](https://www.youtube.com/watch?v=YIVkERhy8xo&t=185s) craft. Uh, so like I said, you have like [3:08](https://www.youtube.com/watch?v=YIVkERhy8xo&t=188s) codecs and uh, open code etc. [3:12](https://www.youtube.com/watch?v=YIVkERhy8xo&t=192s) at your command. [3:13](https://www.youtube.com/watch?v=YIVkERhy8xo&t=193s) >> Uh, and these agents can do everything. [3:16](https://www.youtube.com/watch?v=YIVkERhy8xo&t=196s) They're just normal agents. Uh, so I can [3:18](https://www.youtube.com/watch?v=YIVkERhy8xo&t=198s) just prompt them in this nice side panel [3:20](https://www.youtube.com/watch?v=YIVkERhy8xo&t=200s) that also has voice. So there's [3:23](https://www.youtube.com/watch?v=YIVkERhy8xo&t=203s) multimodality here. Uh, and just do [3:26](https://www.youtube.com/watch?v=YIVkERhy8xo&t=206s) something for me. [3:27](https://www.youtube.com/watch?v=YIVkERhy8xo&t=207s) >> Prompt. Prompt. [3:29](https://www.youtube.com/watch?v=YIVkERhy8xo&t=209s) >> And now these agents can do whatever. [3:33](https://www.youtube.com/watch?v=YIVkERhy8xo&t=213s) Uh, and if you look around you would see [3:35](https://www.youtube.com/watch?v=YIVkERhy8xo&t=215s) that the map has a a bunch of different [3:37](https://www.youtube.com/watch?v=YIVkERhy8xo&t=217s) elements. So you have buildings. These [3:38](https://www.youtube.com/watch?v=YIVkERhy8xo&t=218s) buildings represent functionalities. So [3:40](https://www.youtube.com/watch?v=YIVkERhy8xo&t=220s) for example, if you can manage your [3:42](https://www.youtube.com/watch?v=YIVkERhy8xo&t=222s) plugins or skills, uh you have like [3:44](https://www.youtube.com/watch?v=YIVkERhy8xo&t=224s) integrated terminal, integrated git, uh [3:47](https://www.youtube.com/watch?v=YIVkERhy8xo&t=227s) you can do in your all of your work from [3:50](https://www.youtube.com/watch?v=YIVkERhy8xo&t=230s) within this space. [3:52](https://www.youtube.com/watch?v=YIVkERhy8xo&t=232s) So now that we have this and we [3:54](https://www.youtube.com/watch?v=YIVkERhy8xo&t=234s) understand the basics, let's look at [3:56](https://www.youtube.com/watch?v=YIVkERhy8xo&t=236s) what we can do uh to raise the ceiling. [3:59](https://www.youtube.com/watch?v=YIVkERhy8xo&t=239s) So the first step is obviously [4:00](https://www.youtube.com/watch?v=YIVkERhy8xo&t=240s) visibility. I need to understand what's [4:02](https://www.youtube.com/watch?v=YIVkERhy8xo&t=242s) going on. What are my agents doing? So, [4:05](https://www.youtube.com/watch?v=YIVkERhy8xo&t=245s) I have this nice side panel and it shows [4:07](https://www.youtube.com/watch?v=YIVkERhy8xo&t=247s) me the different agents and what their [4:09](https://www.youtube.com/watch?v=YIVkERhy8xo&t=249s) task is and what the last thing they did [4:10](https://www.youtube.com/watch?v=YIVkERhy8xo&t=250s) is and what they're currently doing. [4:12](https://www.youtube.com/watch?v=YIVkERhy8xo&t=252s) Everything I need to quickly understand [4:14](https://www.youtube.com/watch?v=YIVkERhy8xo&t=254s) who needs my attention and who doesn't. [4:17](https://www.youtube.com/watch?v=YIVkERhy8xo&t=257s) The second part of it is we're not just [4:20](https://www.youtube.com/watch?v=YIVkERhy8xo&t=260s) limited to this. We have a map. So, I [4:22](https://www.youtube.com/watch?v=YIVkERhy8xo&t=262s) can take let's say my file system, [4:25](https://www.youtube.com/watch?v=YIVkERhy8xo&t=265s) project it onto the map, and now I can [4:27](https://www.youtube.com/watch?v=YIVkERhy8xo&t=267s) see exactly what my agents are doing. [4:30](https://www.youtube.com/watch?v=YIVkERhy8xo&t=270s) Not just by looking at a list of stuff. [4:32](https://www.youtube.com/watch?v=YIVkERhy8xo&t=272s) I can actually see okay this agent is [4:34](https://www.youtube.com/watch?v=YIVkERhy8xo&t=274s) now working on this directory or this [4:36](https://www.youtube.com/watch?v=YIVkERhy8xo&t=276s) file and these files are represented as [4:38](https://www.youtube.com/watch?v=YIVkERhy8xo&t=278s) runes on the map. So I can really look [4:41](https://www.youtube.com/watch?v=YIVkERhy8xo&t=281s) at each one and see exactly what each [4:43](https://www.youtube.com/watch?v=YIVkERhy8xo&t=283s) agent did and when. And now I can take [4:46](https://www.youtube.com/watch?v=YIVkERhy8xo&t=286s) this nice information because we are the [4:49](https://www.youtube.com/watch?v=YIVkERhy8xo&t=289s) orchestrator uh and use that to create [4:52](https://www.youtube.com/watch?v=YIVkERhy8xo&t=292s) linkages and now I can create cool stuff [4:54](https://www.youtube.com/watch?v=YIVkERhy8xo&t=294s) like heat maps. So I can really easily [4:57](https://www.youtube.com/watch?v=YIVkERhy8xo&t=297s) see what's going on. And the the other [5:02](https://www.youtube.com/watch?v=YIVkERhy8xo&t=302s) aspect is that now that we have this [5:04](https://www.youtube.com/watch?v=YIVkERhy8xo&t=304s) visibility, that's not really enough. We [5:06](https://www.youtube.com/watch?v=YIVkERhy8xo&t=306s) need to be able to react to what's going [5:08](https://www.youtube.com/watch?v=YIVkERhy8xo&t=308s) on. If someone needs my attention, I [5:10](https://www.youtube.com/watch?v=YIVkERhy8xo&t=310s) need to be able to quickly jump to it. [5:12](https://www.youtube.com/watch?v=YIVkERhy8xo&t=312s) So much like video games or RTS games or [5:16](https://www.youtube.com/watch?v=YIVkERhy8xo&t=316s) Civilization, whatever, you would click [5:18](https://www.youtube.com/watch?v=YIVkERhy8xo&t=318s) uh spacebar and it would just take you [5:20](https://www.youtube.com/watch?v=YIVkERhy8xo&t=320s) to the nearest thing that actually [5:21](https://www.youtube.com/watch?v=YIVkERhy8xo&t=321s) needed your attention. So you can really [5:23](https://www.youtube.com/watch?v=YIVkERhy8xo&t=323s) easily understand what's going on. and [5:26](https://www.youtube.com/watch?v=YIVkERhy8xo&t=326s) jump from one thing to the other and [5:28](https://www.youtube.com/watch?v=YIVkERhy8xo&t=328s) answer questions or approve plans, etc. [5:34](https://www.youtube.com/watch?v=YIVkERhy8xo&t=334s) That was nice. I mean, it's a nice step [5:36](https://www.youtube.com/watch?v=YIVkERhy8xo&t=336s) just to have that visibility, but it's [5:38](https://www.youtube.com/watch?v=YIVkERhy8xo&t=338s) not really enough. It's pretty [5:41](https://www.youtube.com/watch?v=YIVkERhy8xo&t=341s) exhausting to keep looking at what the [5:43](https://www.youtube.com/watch?v=YIVkERhy8xo&t=343s) agents are doing in the granularity of [5:46](https://www.youtube.com/watch?v=YIVkERhy8xo&t=346s) what files are they using and what tools [5:47](https://www.youtube.com/watch?v=YIVkERhy8xo&t=347s) are they using at the moment. So, we [5:49](https://www.youtube.com/watch?v=YIVkERhy8xo&t=349s) need something else. Uh and the reason [5:52](https://www.youtube.com/watch?v=YIVkERhy8xo&t=352s) is that we have like two two main [5:54](https://www.youtube.com/watch?v=YIVkERhy8xo&t=354s) problems here. Uh one is that there's [5:56](https://www.youtube.com/watch?v=YIVkERhy8xo&t=356s) just too much to hold in our heads. Even [5:58](https://www.youtube.com/watch?v=YIVkERhy8xo&t=358s) if I can understand what's going on and [6:00](https://www.youtube.com/watch?v=YIVkERhy8xo&t=360s) I can see everything, I can't really do [6:02](https://www.youtube.com/watch?v=YIVkERhy8xo&t=362s) like 20 things at the same time. I can't [6:04](https://www.youtube.com/watch?v=YIVkERhy8xo&t=364s) even think of the next thing I want to [6:06](https://www.youtube.com/watch?v=YIVkERhy8xo&t=366s) do if I have this, you know, huge coding [6:08](https://www.youtube.com/watch?v=YIVkERhy8xo&t=368s) project or whatever. So I need to find [6:11](https://www.youtube.com/watch?v=YIVkERhy8xo&t=371s) some way to fix that. Uh so we can I [6:13](https://www.youtube.com/watch?v=YIVkERhy8xo&t=373s) mean we can tell the agents uh you know [6:16](https://www.youtube.com/watch?v=YIVkERhy8xo&t=376s) go through my codebase, find out what [6:18](https://www.youtube.com/watch?v=YIVkERhy8xo&t=378s) the next things I need to do are, create [6:20](https://www.youtube.com/watch?v=YIVkERhy8xo&t=380s) these tasks for me and I just click a [6:21](https://www.youtube.com/watch?v=YIVkERhy8xo&t=381s) button, I accept the quest and there's [6:24](https://www.youtube.com/watch?v=YIVkERhy8xo&t=384s) some agent that does it for me which is [6:26](https://www.youtube.com/watch?v=YIVkERhy8xo&t=386s) fun. Uh it's cool. Uh but the next [6:30](https://www.youtube.com/watch?v=YIVkERhy8xo&t=390s) problem here is that okay now I have 20 [6:32](https://www.youtube.com/watch?v=YIVkERhy8xo&t=392s) stuff going at the same time and I have [6:33](https://www.youtube.com/watch?v=YIVkERhy8xo&t=393s) to babysit each and every one and that [6:36](https://www.youtube.com/watch?v=YIVkERhy8xo&t=396s) is completely exhausting. So you could [6:39](https://www.youtube.com/watch?v=YIVkERhy8xo&t=399s) ask okay we have agents agents are [6:41](https://www.youtube.com/watch?v=YIVkERhy8xo&t=401s) amazing as we've established so I can [6:45](https://www.youtube.com/watch?v=YIVkERhy8xo&t=405s) just go ahead and tell the agent to do [6:47](https://www.youtube.com/watch?v=YIVkERhy8xo&t=407s) it for me now it's called software [6:49](https://www.youtube.com/watch?v=YIVkERhy8xo&t=409s) factories but at the time it was just [6:51](https://www.youtube.com/watch?v=YIVkERhy8xo&t=411s) doing stuff autonomously uh so you can [6:53](https://www.youtube.com/watch?v=YIVkERhy8xo&t=413s) give it like a general sense of what you [6:55](https://www.youtube.com/watch?v=YIVkERhy8xo&t=415s) want it would spawn an orchestrator and [6:58](https://www.youtube.com/watch?v=YIVkERhy8xo&t=418s) break it down for you and run it all in [7:00](https://www.youtube.com/watch?v=YIVkERhy8xo&t=420s) well in this case it's a local uh [7:03](https://www.youtube.com/watch?v=YIVkERhy8xo&t=423s) container so everything is just done in [7:05](https://www.youtube.com/watch?v=YIVkERhy8xo&t=425s) isolation and I don't need to do any [7:07](https://www.youtube.com/watch?v=YIVkERhy8xo&t=427s) babysitting the agent does it for me. On [7:10](https://www.youtube.com/watch?v=YIVkERhy8xo&t=430s) top of that, you have loops, which [7:12](https://www.youtube.com/watch?v=YIVkERhy8xo&t=432s) everyone really likes right now. So, you [7:14](https://www.youtube.com/watch?v=YIVkERhy8xo&t=434s) can do cool things like just tell it to, [7:16](https://www.youtube.com/watch?v=YIVkERhy8xo&t=436s) I don't know, scan Twitter for cool [7:18](https://www.youtube.com/watch?v=YIVkERhy8xo&t=438s) stuff you want to build or scan your [7:20](https://www.youtube.com/watch?v=YIVkERhy8xo&t=440s) GitHub, and it will just create these [7:22](https://www.youtube.com/watch?v=YIVkERhy8xo&t=442s) nice uh features for you in the [7:24](https://www.youtube.com/watch?v=YIVkERhy8xo&t=444s) background without you needing to do too [7:26](https://www.youtube.com/watch?v=YIVkERhy8xo&t=446s) much up front. [7:29](https://www.youtube.com/watch?v=YIVkERhy8xo&t=449s) The bad thing is that I mean this is [7:32](https://www.youtube.com/watch?v=YIVkERhy8xo&t=452s) nice uh but what do I do when I have [7:35](https://www.youtube.com/watch?v=YIVkERhy8xo&t=455s) again I don't know I would say 20 but [7:37](https://www.youtube.com/watch?v=YIVkERhy8xo&t=457s) even five things to review at the same [7:40](https://www.youtube.com/watch?v=YIVkERhy8xo&t=460s) time. Uh it's very very hard to get that [7:43](https://www.youtube.com/watch?v=YIVkERhy8xo&t=463s) out of the way. So what I did is create [7:47](https://www.youtube.com/watch?v=YIVkERhy8xo&t=467s) uh this like review kit. So I get the [7:49](https://www.youtube.com/watch?v=YIVkERhy8xo&t=469s) entire changes I need and I get uh what [7:52](https://www.youtube.com/watch?v=YIVkERhy8xo&t=472s) files changed if you want to go like [7:54](https://www.youtube.com/watch?v=YIVkERhy8xo&t=474s) file by file or you can get just get [7:56](https://www.youtube.com/watch?v=YIVkERhy8xo&t=476s) visual evidence. So you can actually see [7:58](https://www.youtube.com/watch?v=YIVkERhy8xo&t=478s) videos of what changed or photos of what [8:00](https://www.youtube.com/watch?v=YIVkERhy8xo&t=480s) changed so you can really quickly [8:02](https://www.youtube.com/watch?v=YIVkERhy8xo&t=482s) understand what you want or even run a [8:04](https://www.youtube.com/watch?v=YIVkERhy8xo&t=484s) few uh instances in parallel and just [8:07](https://www.youtube.com/watch?v=YIVkERhy8xo&t=487s) pick the best implementation out of [8:09](https://www.youtube.com/watch?v=YIVkERhy8xo&t=489s) these. [8:11](https://www.youtube.com/watch?v=YIVkERhy8xo&t=491s) So now that we have this we have uh [8:13](https://www.youtube.com/watch?v=YIVkERhy8xo&t=493s) visibility and we have autonomy uh it's [8:16](https://www.youtube.com/watch?v=YIVkERhy8xo&t=496s) pretty cool like we reached a pretty [8:17](https://www.youtube.com/watch?v=YIVkERhy8xo&t=497s) good place in raising the ceiling of [8:20](https://www.youtube.com/watch?v=YIVkERhy8xo&t=500s) what we can do with agents but it's [8:23](https://www.youtube.com/watch?v=YIVkERhy8xo&t=503s) still pretty exhausting. I mean it's it [8:25](https://www.youtube.com/watch?v=YIVkERhy8xo&t=505s) was fun. Uh, but now I have to ask [8:27](https://www.youtube.com/watch?v=YIVkERhy8xo&t=507s) myself, do I really need to be the only [8:30](https://www.youtube.com/watch?v=YIVkERhy8xo&t=510s) one doing this? It's pretty lonely to [8:32](https://www.youtube.com/watch?v=YIVkERhy8xo&t=512s) just talk to agents all day and trying [8:34](https://www.youtube.com/watch?v=YIVkERhy8xo&t=514s) to and you know, if you look at uh what [8:37](https://www.youtube.com/watch?v=YIVkERhy8xo&t=517s) a team does, for example, if it's just [8:39](https://www.youtube.com/watch?v=YIVkERhy8xo&t=519s) me, that's fine. But if I have a team, I [8:42](https://www.youtube.com/watch?v=YIVkERhy8xo&t=522s) might not even be the right person to [8:44](https://www.youtube.com/watch?v=YIVkERhy8xo&t=524s) review what's going on. So, I need [8:46](https://www.youtube.com/watch?v=YIVkERhy8xo&t=526s) something else. Uh, and that something [8:48](https://www.youtube.com/watch?v=YIVkERhy8xo&t=528s) is collaboration. We can just have other [8:52](https://www.youtube.com/watch?v=YIVkERhy8xo&t=532s) people join us. So the way that looks in [8:56](https://www.youtube.com/watch?v=YIVkERhy8xo&t=536s) agent craft is that I can create these [8:59](https://www.youtube.com/watch?v=YIVkERhy8xo&t=539s) uh you can call it you know war halls [9:01](https://www.youtube.com/watch?v=YIVkERhy8xo&t=541s) war room whatever uh host them it's like [9:04](https://www.youtube.com/watch?v=YIVkERhy8xo&t=544s) completely local but you can do like [9:06](https://www.youtube.com/watch?v=YIVkERhy8xo&t=546s) tunnels and people can join you h so I [9:09](https://www.youtube.com/watch?v=YIVkERhy8xo&t=549s) can create a room and let's say I have a [9:11](https://www.youtube.com/watch?v=YIVkERhy8xo&t=551s) product designer in my team in this case [9:13](https://www.youtube.com/watch?v=YIVkERhy8xo&t=553s) uh it's my wife and I can approve uh her [9:16](https://www.youtube.com/watch?v=YIVkERhy8xo&t=556s) otherwise she'll be mad uh so I can [9:18](https://www.youtube.com/watch?v=YIVkERhy8xo&t=558s) approve her and she does her thing she [9:21](https://www.youtube.com/watch?v=YIVkERhy8xo&t=561s) designs whatever and I as an engineer [9:24](https://www.youtube.com/watch?v=YIVkERhy8xo&t=564s) can not only see that that's that's her [9:28](https://www.youtube.com/watch?v=YIVkERhy8xo&t=568s) um I can not only see that on my map uh [9:31](https://www.youtube.com/watch?v=YIVkERhy8xo&t=571s) I can actually take that and follow off [9:34](https://www.youtube.com/watch?v=YIVkERhy8xo&t=574s) off of it [9:35](https://www.youtube.com/watch?v=YIVkERhy8xo&t=575s) >> prompt prompt [9:36](https://www.youtube.com/watch?v=YIVkERhy8xo&t=576s) >> yeah that's just like her um so I can [9:41](https://www.youtube.com/watch?v=YIVkERhy8xo&t=581s) take that design follow up on it [9:44](https://www.youtube.com/watch?v=YIVkERhy8xo&t=584s) implement it while she keeps working uh [9:46](https://www.youtube.com/watch?v=YIVkERhy8xo&t=586s) and the cool thing here is that not only [9:48](https://www.youtube.com/watch?v=YIVkERhy8xo&t=588s) can I hand off or fork and kind of have [9:50](https://www.youtube.com/watch?v=YIVkERhy8xo&t=590s) this shared workspace so I'm not really [9:52](https://www.youtube.com/watch?v=YIVkERhy8xo&t=592s) bound anymore or to Git. I can use Git, [9:54](https://www.youtube.com/watch?v=YIVkERhy8xo&t=594s) but I'm not really bound to something. [9:56](https://www.youtube.com/watch?v=YIVkERhy8xo&t=596s) Everyone is working together. Uh you can [9:58](https://www.youtube.com/watch?v=YIVkERhy8xo&t=598s) see that there's also some kind of chat [10:00](https://www.youtube.com/watch?v=YIVkERhy8xo&t=600s) between not only between the people [10:03](https://www.youtube.com/watch?v=YIVkERhy8xo&t=603s) within the room, but also between [10:05](https://www.youtube.com/watch?v=YIVkERhy8xo&t=605s) agents. So we can have this really nice [10:08](https://www.youtube.com/watch?v=YIVkERhy8xo&t=608s) notice board of exactly what each person [10:10](https://www.youtube.com/watch?v=YIVkERhy8xo&t=610s) is doing, what each agent doing. They're [10:11](https://www.youtube.com/watch?v=YIVkERhy8xo&t=611s) all aware of each other and they can [10:13](https://www.youtube.com/watch?v=YIVkERhy8xo&t=613s) collaborate much more closely uh than [10:15](https://www.youtube.com/watch?v=YIVkERhy8xo&t=615s) you normally would be able to do. Uh so [10:17](https://www.youtube.com/watch?v=YIVkERhy8xo&t=617s) for example, here I can say, okay, I [10:19](https://www.youtube.com/watch?v=YIVkERhy8xo&t=619s) took off this design task. [10:22](https://www.youtube.com/watch?v=YIVkERhy8xo&t=622s) Uh so now we have oh by the way this is [10:26](https://www.youtube.com/watch?v=YIVkERhy8xo&t=626s) also from devices you can have like [10:28](https://www.youtube.com/watch?v=YIVkERhy8xo&t=628s) mobile and and if you like telegram you [10:30](https://www.youtube.com/watch?v=YIVkERhy8xo&t=630s) can use telegram whatever floats your [10:33](https://www.youtube.com/watch?v=YIVkERhy8xo&t=633s) boat. So we have visibility and we have [10:36](https://www.youtube.com/watch?v=YIVkERhy8xo&t=636s) autonomy and we have collaboration so we [10:38](https://www.youtube.com/watch?v=YIVkERhy8xo&t=638s) managed to raise the ceiling. Uh it's [10:40](https://www.youtube.com/watch?v=YIVkERhy8xo&t=640s) now possible especially if you're power [10:43](https://www.youtube.com/watch?v=YIVkERhy8xo&t=643s) users to do really cool stuff and use [10:45](https://www.youtube.com/watch?v=YIVkERhy8xo&t=645s) more agents and do more things. But the [10:48](https://www.youtube.com/watch?v=YIVkERhy8xo&t=648s) interesting thing here was, you know, if [10:50](https://www.youtube.com/watch?v=YIVkERhy8xo&t=650s) I looked at the feedback I got, it's not [10:53](https://www.youtube.com/watch?v=YIVkERhy8xo&t=653s) just that power users were able to do [10:55](https://www.youtube.com/watch?v=YIVkERhy8xo&t=655s) more. The cooler thing for me was that [10:57](https://www.youtube.com/watch?v=YIVkERhy8xo&t=657s) people that had even nothing to do with [10:59](https://www.youtube.com/watch?v=YIVkERhy8xo&t=659s) agents or nothing to do with [11:00](https://www.youtube.com/watch?v=YIVkERhy8xo&t=660s) productivity, uh, come and say, uh, [11:03](https://www.youtube.com/watch?v=YIVkERhy8xo&t=663s) okay, my kids really loved it and they [11:06](https://www.youtube.com/watch?v=YIVkERhy8xo&t=666s) use it all the time to to orchestrate [11:08](https://www.youtube.com/watch?v=YIVkERhy8xo&t=668s) agents, which they didn't do until now. [11:10](https://www.youtube.com/watch?v=YIVkERhy8xo&t=670s) uh or you know someone that flunked out [11:13](https://www.youtube.com/watch?v=YIVkERhy8xo&t=673s) of college because he played Starcraft [11:15](https://www.youtube.com/watch?v=YIVkERhy8xo&t=675s) and now you can suddenly do really [11:16](https://www.youtube.com/watch?v=YIVkERhy8xo&t=676s) productive stuff with it. Uh and a bunch [11:19](https://www.youtube.com/watch?v=YIVkERhy8xo&t=679s) of other examples of kind of bringing [11:21](https://www.youtube.com/watch?v=YIVkERhy8xo&t=681s) communities that so far were kind of [11:23](https://www.youtube.com/watch?v=YIVkERhy8xo&t=683s) behind into this new world of [11:25](https://www.youtube.com/watch?v=YIVkERhy8xo&t=685s) productivity. [11:27](https://www.youtube.com/watch?v=YIVkERhy8xo&t=687s) So we need to raise the ceiling but we [11:30](https://www.youtube.com/watch?v=YIVkERhy8xo&t=690s) also need to uh lower the floor. We need [11:34](https://www.youtube.com/watch?v=YIVkERhy8xo&t=694s) to have some kind of way to take no [11:37](https://www.youtube.com/watch?v=YIVkERhy8xo&t=697s) probably around 90% of the people in the [11:39](https://www.youtube.com/watch?v=YIVkERhy8xo&t=699s) world and bring them into this hygienic [11:41](https://www.youtube.com/watch?v=YIVkERhy8xo&t=701s) future without having them burn out. [11:45](https://www.youtube.com/watch?v=YIVkERhy8xo&t=705s) So again, I took it too far. Um, and [11:48](https://www.youtube.com/watch?v=YIVkERhy8xo&t=708s) there's this uh currently named TBD um [11:53](https://www.youtube.com/watch?v=YIVkERhy8xo&t=713s) potentially loopers. Uh the idea here is [11:55](https://www.youtube.com/watch?v=YIVkERhy8xo&t=715s) okay why don't why do we need something [11:57](https://www.youtube.com/watch?v=YIVkERhy8xo&t=717s) like you know Warcraft which is [11:59](https://www.youtube.com/watch?v=YIVkERhy8xo&t=719s) potentially very complicated uh [12:01](https://www.youtube.com/watch?v=YIVkERhy8xo&t=721s) especially or might be even daunting if [12:03](https://www.youtube.com/watch?v=YIVkERhy8xo&t=723s) you're not into the granularity of what [12:05](https://www.youtube.com/watch?v=YIVkERhy8xo&t=725s) files are there are or how uh complex [12:07](https://www.youtube.com/watch?v=YIVkERhy8xo&t=727s) the UI is and kind of try to find a [12:10](https://www.youtube.com/watch?v=YIVkERhy8xo&t=730s) lower denominator of what makes stuff [12:12](https://www.youtube.com/watch?v=YIVkERhy8xo&t=732s) accessible especially now that agents [12:15](https://www.youtube.com/watch?v=YIVkERhy8xo&t=735s) are much more capable so instead of [12:17](https://www.youtube.com/watch?v=YIVkERhy8xo&t=737s) granularity or having like exactly what [12:20](https://www.youtube.com/watch?v=YIVkERhy8xo&t=740s) files they're working on maybe I can [12:22](https://www.youtube.com/watch?v=YIVkERhy8xo&t=742s) have it more of a mobile game level uh [12:25](https://www.youtube.com/watch?v=YIVkERhy8xo&t=745s) have projects that they keep returning [12:27](https://www.youtube.com/watch?v=YIVkERhy8xo&t=747s) to and use loops and other stuff that [12:31](https://www.youtube.com/watch?v=YIVkERhy8xo&t=751s) rely on agents being more autonomous to [12:33](https://www.youtube.com/watch?v=YIVkERhy8xo&t=753s) have like a very basic level of I prompt [12:36](https://www.youtube.com/watch?v=YIVkERhy8xo&t=756s) I get the result I don't really care [12:38](https://www.youtube.com/watch?v=YIVkERhy8xo&t=758s) perhaps about the way I can see it if I [12:40](https://www.youtube.com/watch?v=YIVkERhy8xo&t=760s) want to and I have this really nice [12:42](https://www.youtube.com/watch?v=YIVkERhy8xo&t=762s) interactive and cool experience that [12:44](https://www.youtube.com/watch?v=YIVkERhy8xo&t=764s) makes me want to come back to this and [12:47](https://www.youtube.com/watch?v=YIVkERhy8xo&t=767s) obviously I mean this is one [12:49](https://www.youtube.com/watch?v=YIVkERhy8xo&t=769s) visualization uh it can really resonate [12:53](https://www.youtube.com/watch?v=YIVkERhy8xo&t=773s) with some people but might not might not [12:55](https://www.youtube.com/watch?v=YIVkERhy8xo&t=775s) present to others because we're not all [12:57](https://www.youtube.com/watch?v=YIVkERhy8xo&t=777s) the same and what may appear like [13:00](https://www.youtube.com/watch?v=YIVkERhy8xo&t=780s) amazing and easy to one person can [13:01](https://www.youtube.com/watch?v=YIVkERhy8xo&t=781s) appear very difficult to another. Uh but [13:05](https://www.youtube.com/watch?v=YIVkERhy8xo&t=785s) here uh I'm trying to look at a way to [13:08](https://www.youtube.com/watch?v=YIVkERhy8xo&t=788s) make this more customizable. Try and [13:10](https://www.youtube.com/watch?v=YIVkERhy8xo&t=790s) understand what it is that you as a [13:12](https://www.youtube.com/watch?v=YIVkERhy8xo&t=792s) person need in order to understand and [13:15](https://www.youtube.com/watch?v=YIVkERhy8xo&t=795s) want to use these tools. So this is a [13:19](https://www.youtube.com/watch?v=YIVkERhy8xo&t=799s) very very early draft. Um but if you [13:22](https://www.youtube.com/watch?v=YIVkERhy8xo&t=802s) want to use it, please let me know. So [13:24](https://www.youtube.com/watch?v=YIVkERhy8xo&t=804s) we're the bottleneck. I think we've [13:26](https://www.youtube.com/watch?v=YIVkERhy8xo&t=806s) established that uh pretty well until [13:28](https://www.youtube.com/watch?v=YIVkERhy8xo&t=808s) now. But the skills that we need to stop [13:30](https://www.youtube.com/watch?v=YIVkERhy8xo&t=810s) being the bottleneck and enable this new [13:32](https://www.youtube.com/watch?v=YIVkERhy8xo&t=812s) future are already with us. So we are [13:35](https://www.youtube.com/watch?v=YIVkERhy8xo&t=815s) the bottleneck, but we don't have to be. [13:39](https://www.youtube.com/watch?v=YIVkERhy8xo&t=819s) And I hope that you all can take what's [13:42](https://www.youtube.com/watch?v=YIVkERhy8xo&t=822s) relevant to you and your life and apply [13:44](https://www.youtube.com/watch?v=YIVkERhy8xo&t=824s) it forward. [13:46](https://www.youtube.com/watch?v=YIVkERhy8xo&t=826s) So I would ask you for two things. One, [13:49](https://www.youtube.com/watch?v=YIVkERhy8xo&t=829s) agent is already out there. Like you can [13:50](https://www.youtube.com/watch?v=YIVkERhy8xo&t=830s) just npx install agent as a website uh [13:54](https://www.youtube.com/watch?v=YIVkERhy8xo&t=834s) and do your thing. Uh this new thing [13:56](https://www.youtube.com/watch?v=YIVkERhy8xo&t=836s) that I just showed you uh potentially [13:59](https://www.youtube.com/watch?v=YIVkERhy8xo&t=839s) loopers uh is still experimental. I'm [14:02](https://www.youtube.com/watch?v=YIVkERhy8xo&t=842s) still trying to understand the interest [14:05](https://www.youtube.com/watch?v=YIVkERhy8xo&t=845s) and what people are actually looking for [14:07](https://www.youtube.com/watch?v=YIVkERhy8xo&t=847s) to uh get the floor that they want or [14:09](https://www.youtube.com/watch?v=YIVkERhy8xo&t=849s) the ceiling that they want. So, please [14:12](https://www.youtube.com/watch?v=YIVkERhy8xo&t=852s) uh go here and express your interest and [14:14](https://www.youtube.com/watch?v=YIVkERhy8xo&t=854s) tell me what you're looking for so I can [14:16](https://www.youtube.com/watch?v=YIVkERhy8xo&t=856s) know what to do with this. And thank [14:19](https://www.youtube.com/watch?v=YIVkERhy8xo&t=859s) you. [14:33](https://www.youtube.com/watch?v=YIVkERhy8xo&t=873s) I got it.