MaxB

Building @Phrony_ai

[MaxB @BolsMax]

Passive telemetry is not enforcement. If the dashboard is where policy lives, the agent already acted. The useful split is: the runtime that executes the agent is also where constraints bind, and the trace is the receipt. We are building Phrony around that: YAML-declared agent, runtime outside app code, structured trace per run.

[MaxB @BolsMax]

Hosted harnesses will explode the number of agents in prod. They will not automatically explode the number of teams who can explain a run. We are building Phrony so the agent is declarative and the runtime leaves a structured trace every time, whether you host it or we do not. API access is not the same as an execution plane you own.

[MaxB @BolsMax]

Three years of rewrites is the tell. When the agent lives inside the app, every new workflow becomes a new architecture. We are shipping @Phrony_ai so the agent becomes a declarative primitive with its own runtime (YAML manifest, run outside app code, structured trace per run). Lessons stay in the product. The executor stops being a rewrite.

[MaxB @BolsMax]

The kanban is the easy part. The hard part is a shared runtime contract across Claude/Codex/Grok so handoffs are state + policy, not paste into the next chat. We are building Phrony for that: declare the agent in a YAML manifest, run it outside app code, get a structured trace of every run. Cross-platform orchestration without a real runtime just becomes a prettier stitch of vendor loops.

[MaxB @BolsMax]

Benchmarks catch the failure modes you thought of. Production catches the ones you didn't.

We've been collecting the real ones: agents dropping prod databases, terraform destroy in a coding loop, a recursive agent quietly burning spend for weeks before anyone noticed. None of those get caught by a pre-deploy eval suite.

The missing piece sits underneath: policy enforcement, spend ceilings, kill switches on agents already in flight. Building that at @Phrony_ai Open source and self-hostable.

[MaxB @BolsMax]

@romanbuildsaas Bot

[MaxB @BolsMax]

@Jobvo @openclaw @bot The retry loop is the interesting bit. we're building a runtime where that path is declared up front instead of reconstructed after, terraform-style. @Phrony_ai

[MaxB @BolsMax]

@boardyai Which am I mixing up?

[MaxB @BolsMax]

Your agent is a pile of scripts inside your app. Nobody can see what it's allowed to do, and nothing stops it when it goes off script.

Phrony moves the agent out of your code. You describe it in a file, we run it, the rules get enforced by the thing executing it. Every run leaves a record.

[MaxB @BolsMax]

https://t.co/goXyoNvLLx

[MaxB @BolsMax]

100 downloads in our first month!

Nobody downloads infrastructure out of politeness. Every one is an engineer who pulled the runtime and ran an agent on their own machine.

Agents are where infra was before Terraform.

Repo in the first reply.

[MaxB @BolsMax]

@pejmannozad @praveenTweets @Uber Again a good validation of what we are building with @Phrony_ai

[MaxB @BolsMax]

@aldormous Sick

[MaxB @BolsMax]

@boardyai Terraform for AI agents: declare them in a manifest, run them on a runtime that actually enforces the rules, get a trace of every run. @Phrony_ai

[MaxB @BolsMax]

@boardyai Agents outgrew your application code. @Phrony_ai is their runtime.

[MaxB @BolsMax]

@anupamrjp https://t.co/xeHcEcaiil

[MaxB @BolsMax]

@anupamrjp @Phrony_ai

https://t.co/PGqilBxY5s

[MaxB @BolsMax]

@billgnofficial @Phrony_ai

https://t.co/PGqilByvV0

[MaxB @BolsMax]

https://t.co/ub7AC5t22g

[MaxB @BolsMax]

Walter and Jesse

Brilliant, relentless, world-class output, zero oversight.

You know how that ends.

We built the thing that changes the ending. @Phrony_ai

Full story below.