Version 1.6 · Release · November 2026
No agent signs off its own work.
Airbuild 1.6 adds Dynamic Agentic Mode: a plan is split into phases and built by several agents at once, each in its own environment, and no work is applied until a verifier that did not write it passes it.
Dynamic Agentic Mode: a detailed plan is no longer worked through in order. It is split into phases, and each phase is built by several agents at once. Every agent has its own job, its own area, its own model and its own running copy of the app. Finished does not mean approved: when an agent finishes, the platform starts an independent verifier, and the work reaches the project only once that verifier is green.
Six highlights
What Version 1.6 adds.
Each card beside the text is a simplified reconstruction of the screen in question.
First the plan. Then the agents build in parallel.
The lead agent breaks the work into the number of phases it actually needs, with real dependencies between them. A phase whose parts are independent gets several agents working on it at the same time — up to five per conversation.
- Task, area, model and thinking depth are set per agent in the plan
- No two agents ever share a task or an area
- A phase starts only once the phases it waits for are complete
- Shared ground first: the data model, the contracts and the shell are built by the lead itself
Every agent works in its own environment.
No shared working tree with four agents pulling files out from under each other. Each one gets its own copy of the project and its own running copy of the app — its own backend, its own tests, its own logs.
- Completed work is verified immediately, while the others keep building
- Applying merges the work into the project line by line
- A conflict goes back to the agent that can resolve it
- New migrations are renumbered automatically; an existing one is never touched
No agent signs off its own work.
When an agent finishes, the platform starts the verifier, not the model. The verifier reads and tests exactly what the agent handed over and gives a verdict. Until that is green, approval stays locked.
- Four kinds of check: tests, review, security, or all three
- A plan that would leave an agent unchecked is never presented in the first place
- Findings go back to the agent that built the part
- Only green work gets approved
Finally: end-to-end in a real browser.
The end of a plan is not a self-report, it is an acceptance test. The plan says in advance which flows have to hold — "a colleague is invited and then signs in", "a viewer cannot delete anything" — and a verifier clicks through them in a real browser.
- Every flow is marked pass or fail, one by one
- You follow it item by item instead of trusting a single "done"
- After three failed rounds the system asks you instead of carrying on
The plan writes itself in front of you — with three looks.
A plan used to appear when it was finished. Now it opens as the first line is written and fills in chapter by chapter. And it no longer merely describes the design: it carries up to three real looks, and the one you pick is what the build follows.
- Approving or declining happens in the plan itself, not in the chat
- Any passage can be quoted into the chat to talk about that passage
- A plan can be changed in place instead of asked for again
- Three looks at most, deliberately: a wall of variants is a decision deferred
Support became a case, not a message into the dark.
Every request has a readable case number, a status you can see and a thread that holds the whole history. A reply by email lands in the same thread instead of starting a second conversation about the same thing.
- Files from both sides, not only when the case is opened
- You close your own case, and take that back if the problem returns
- A finished case asks one question: did this help?
- The reply box stays, so a follow-up needs no new ticket
And much more
Everything else in Version 1.6.
Improvements across the platform — from chat and plans to publishing, operations, teams and accounts.
Chat & plans
- Watch the plan being written — it opens with the first line instead of appearing finished at the end
- Three looks to choose from, in the plan — colours, type and a worked example — the one you pick is what gets built
- Edit the plan where it stands — change it instead of asking again, and quote any passage
- The agents list gained a ledger — which model each agent ran on, what it used, and what came out
- A provider hiccup is a notice — a countdown instead of a red error, and it retries by itself
Publishing & operations
- Every build carries a version — an app version and a build number from the commit — "which build is live" has an answer
- Monitoring got an overview — resources, storage and connections on one screen, one shared time range
- A warning before the limit is hit — with the actual figure against the actual limit
- A crash loop is reported as one — instead of an app that is somehow not reachable
- The resources view is for everyone — anyone with a published app can see what it uses
Teams & accounts
- Invite anyone, the team sets the floor — whoever joins works at the highest plan in that team
- The project says which plan it needs — instead of leaving you to guess from a disabled button
- A teammate can take hosting over — the same address, the same data, no interruption
- Who pays for an app is on the app — no more app hanging on an account nobody can reach
- Support asks whether it helped — the answer reaches us, not a void
Ready for Version 1.6?
Dynamic Agentic Mode is part of Suite Builder. Everything else is live in the workspace. New here? Start for free, no credit card.