Live LabsFree Live LabEngineersAgents at work1 hour
Build a Mission Control for AI Agents
An agent that says "refunded, all done" while the double charge is still on the ledger is worse than no agent. Live, Param builds Mission Control for Acme's billing with the OpenAI Agents SDK: a manager hands a double-charge card to a support resolver, the refund waits for a human on Telegram, and the card reaches Done only when the ledger shows one charge per order. An engineering analyst answers whether it is a known bug, and that card closes only on a source it read. The board shows the cost per outcome, failed runs included.
Why this matters
Once an agent works in a chat window, the next request is to let it do real work: pick up issues, update docs, chase approvals. That is where most teams lose the thread. Agents start acting in GitHub and Linear with nobody able to say which agent did what, who allowed it or what it cost. A shared board fixes that the same way it fixes it for people: work is visible, owned and finished in the open.
What happens in the hour
The problem4 min
Why an agent that marks its own work done, or approves its own refund, cannot be trusted with money.
Built live40 min
The board with Backlog, Needs you, In flight and Done; a manager that hands work to the employee hired for it; an approval rule in code that no agent can pass; the ledger as the only judge of Done; the cost of every run.
Questions16 min
How to map the board onto your own company's system of record, and which action to put behind a human first.
What you will be able to do
Build the board
Backlog, Needs you, In flight and Done, with the employee and the cost on every card.
Put a human on the irreversible step
A refund waits for a tap on Telegram, and no AI employee can approve it, the manager included.
Let the system of record decide Done
A card closes only when the ledger agrees, never because an agent said so.
The ideas behind it
A board as the source of truth
Every task an agent takes sits in one column at a time: Backlog, Needs you, In flight or Done. Anyone on the team can see what is running without reading logs.
Roles, not one agent
Each agent has one job, its own instructions and only the tools that job needs, so a mistake stays inside one role.
Approval before the irreversible
An agent pauses and asks a person before it merges, sends or deletes. Everything reversible runs on its own.
Cost on every run
Each run logs its tokens and price, so the monthly bill is a sum you can explain, not a surprise.
Where it fits
The parts of a production agent system, in the order the buildcamp builds them. The lit tiles are the ones this live lab builds; the Agentic AI Buildcamp for Engineers & PMs builds all of them.
Week 1Architect
Roles and modelsOne job per agent, a model chosen for that job, and a token budget.
OrchestrationA queue agents pick work from, with hand-offs a person can follow.
Context and memoryRetrieval with sources, memory across sessions, prompts laid out for the cache.
Tool callingTyped tools that act in GitHub, Linear and Google.
Week 2Build
MCP serversEvery tool behind an MCP gateway, scoped to the role that needs it.
Durable executionRuns that resume after a crash and never repeat a write.
Human in the loopA person approves anything that cannot be undone.
Week 3Secure and deploy
Identity and permissionsEach agent signs in as itself and acts on behalf of a user.
LLM gatewayRouting, caching and a budget on every model call.
Traces, evals and costEvery run traced, scored and charged to the agent that made it.
Multi-tenancyEach team or customer kept apart, in data and in the bill.
Week 4Govern and extend
Policy as codeRules checked on every action, not written in a document.
Signed skillsNew roles built from reviewed, signed skills.
Before you come
For
Engineers who have built an agent and want it doing real work for their team, and the tech leads designing how agents fit into their company's tools.
You need
None. Code is shown, not required.
Track
Agentic AI for engineers who build and run it.
Questions
Do I need to know Claude Code to follow along?
No. The build is shown step by step and the code is explained as it is written. Engineers who use Claude Code will be able to repeat it the same day.
Does this only work with GitHub and Linear?
No. GitHub and Linear are the examples because most teams use them. The last part of the hour shows how to swap the connectors for the tools your team already has.
How is this different from an agent framework?
A framework runs an agent. This lab is about the layer around agents: who owns a task, where a person approves and what each run costs. It works with whichever framework you use.
Is it safe to let agents act in our tools?
Only with limits, which is the point of the lab: scoped tools per role and a human approval on anything that cannot be undone.
Is the Live Lab really free?
Yes. Live Labs are free on Maven. You sign up with your email and get the join link and the recording.
What if I cannot make it live?
Sign up anyway. Everyone who signs up gets the recording, so you can watch the build later and reply with questions.
Do I need to code along?
No. Most people watch the build and ask questions. Every step is shown, so you can repeat it on your own afterwards.