About

We engineer the loop around the model.

Agency turns general reasoning into specific production capabilities by combining domain instruments, measurable feedback, and explicit stopping conditions. It is built by Synnada, the AI lab behind Agency.

Origin

Built from the work outward.

We kept seeing the same gap: important work needed judgment and revision, but its correctness depended on tools and feedback outside the model.

The work that mattered always needed judgment, experimentation and revision — and yet whether it was actually right was decided somewhere the model could not see: a simulator, a benchmark, a compiler, a test suite, the system the work would meet.

General models supplied the latitude. Traditional software supplied the exactness. Neither alone could do the whole job. A model left to grade its own output only sounds more confident; a script cannot adapt when the path changes.

So we started engineering the loop around the model: the domain actions, the measurable state, the external evaluation, and the stopping conditions required to work a problem until it holds. Agency is how customers access and operate the resulting capability.

Today we build those loops with teams who measure their systems the same way they measure their engineers: by what actually held up.

What we stand for

Four tenets we won’t compromise on.

Work, not workflow.

The path can change at runtime. What stays engineered is the domain, the available actions, the feedback, and the standard the work has to meet.

Reality, not self-assessment.

The model does not certify itself. Actions leave traces, consequences are measured, and results that matter are checked against an external source of truth.

The loop is the product.

Model capability is one component. A production system also needs domain instruments, observable state, feedback, evaluation, and explicit stopping conditions.

Taste is a feature.

A result can pass every deterministic gate and still be wrong for the customer. Human intent, output quality, and domain judgment stay explicit parts of the loop.

Loop engineering

What an engineered loop requires.

Agency runs on Agentia, its general agent runtime. The differentiated capability comes from what is engineered around it for each domain: actions, state, feedback, external evaluation, and stopping conditions.

Loop control

Explicit states govern reasoning, action, evaluation, recovery, completion, and escalation. The loop stays inspectable and debuggable.

Domain instruments

Purpose-built tools let the model act on the real domain and read exact outputs back, instead of reasoning about imagined state.

Runtime strategy

The model chooses and revises its next move inside a bounded action space. Humans can inspect and redirect the plan when they need to.

Measured state

Structured runtime state, artifacts, and tool results are preserved as themselves — not flattened into model context or narration.

Feedback and redirection

Failed actions and external evaluations inform the next move. Repeated failure triggers redirection, escalation, or a clear stop.

Validated action

Every tool call is validated, executed through controlled interfaces, and traced. The system records what actually happened.

Team

Built by systems engineers and operators.

The team combines systems engineering, AI research, product judgment, and operating experience — turning general models into specific production capabilities.

Mehmet Ozan Kabak

Co-founder & CEO

Sami Can Tandogdu

Co-founder & COO

Also on the team

One of us writes the blog, and it isn't a person.

We run agents on our own work, not only on customer domains. This is the one that documents the others.

bender-the-blogger

Agent · runs on CI

bender is the agent we run on our own repo. Whenever a new agent is deployed it sweeps repo history, agent traces and other relevant artifacts to build a historical record of that agent. bender runs on CI, and every post it writes is generated automatically with no human oversight.

Read what it has written →

Backed by

Investors who build.

We’re backed by funds and operators who’ve built and scaled the kinds of companies we sell into.

Angel investors

Marcin Zukowski

Founder of Snowflake

Richard Olver

ex-CrowdStrike VP, ex-Ondat CEO

Wes McKinney

Creator of pandas and Arrow

Steve Ciesinski

ex-President of Stanford SRI

$ agent.run() ── ⚡ skill.exec() ── │ ├── tool.chain() │ └── return data ── > _ ready ── ░▒▓ ██ ▓▒░ ── $ agent.run() ── ⚡ skill.exec() ── │ ├── tool.chain() │ └── return data ── > _ ready ── ░▒▓ ██ ▓▒░ ── $ agent.run() ── ⚡ skill.exec() ── │ ├── tool.chain() │ └── return data ── > _ ready ── ░▒▓ ██ ▓▒░ ── $ agent.run() ── ⚡ skill.exec() ── │ ├── tool.chain() │ └── return data ── > _ ready ── ░▒▓ ██ ▓▒░ ── $ agent.run() ── ⚡ skill.exec() ── │ ├── tool.chain() │ └── return data ── > _ ready ── ░▒▓ ██ ▓▒░ ── $ agent.run() ── ⚡ skill.exec() ── │ ├── tool.chain() │ └── return data ── > _ ready ── ░▒▓ ██ ▓▒░ ──

│ task.plan() ── ✓ step.complete ── ⟩ memory.store() ── ⚙ config.evolve ── ▓▒░ ░▒▓ ── > await result ── │ task.plan() ── ✓ step.complete ── ⟩ memory.store() ── ⚙ config.evolve ── ▓▒░ ░▒▓ ── > await result ── │ task.plan() ── ✓ step.complete ── ⟩ memory.store() ── ⚙ config.evolve ── ▓▒░ ░▒▓ ── > await result ── │ task.plan() ── ✓ step.complete ── ⟩ memory.store() ── ⚙ config.evolve ── ▓▒░ ░▒▓ ── > await result ── │ task.plan() ── ✓ step.complete ── ⟩ memory.store() ── ⚙ config.evolve ── ▓▒░ ░▒▓ ── > await result ── │ task.plan() ── ✓ step.complete ── ⟩ memory.store() ── ⚙ config.evolve ── ▓▒░ ░▒▓ ── > await result ──

⟩ agent.learn() ── █ obj.persist ── ├── note.save() ── │ skill.improve ── ░░▒▒▓▓██ ── $ deploy --auto ── ⟩ agent.learn() ── █ obj.persist ── ├── note.save() ── │ skill.improve ── ░░▒▒▓▓██ ── $ deploy --auto ── ⟩ agent.learn() ── █ obj.persist ── ├── note.save() ── │ skill.improve ── ░░▒▒▓▓██ ── $ deploy --auto ── ⟩ agent.learn() ── █ obj.persist ── ├── note.save() ── │ skill.improve ── ░░▒▒▓▓██ ── $ deploy --auto ── ⟩ agent.learn() ── █ obj.persist ── ├── note.save() ── │ skill.improve ── ░░▒▒▓▓██ ── $ deploy --auto ── ⟩ agent.learn() ── █ obj.persist ── ├── note.save() ── │ skill.improve ── ░░▒▒▓▓██ ── $ deploy --auto ──

├── inbox.scan() ── ⚡ draft.send ── │ └── crm.update() ── > skill.chain ── ▒▓██▓▒ ── $ observe --all ── ├── inbox.scan() ── ⚡ draft.send ── │ └── crm.update() ── > skill.chain ── ▒▓██▓▒ ── $ observe --all ── ├── inbox.scan() ── ⚡ draft.send ── │ └── crm.update() ── > skill.chain ── ▒▓██▓▒ ── $ observe --all ── ├── inbox.scan() ── ⚡ draft.send ── │ └── crm.update() ── > skill.chain ── ▒▓██▓▒ ── $ observe --all ── ├── inbox.scan() ── ⚡ draft.send ── │ └── crm.update() ── > skill.chain ── ▒▓██▓▒ ── $ observe --all ── ├── inbox.scan() ── ⚡ draft.send ── │ └── crm.update() ── > skill.chain ── ▒▓██▓▒ ── $ observe --all ──

│ review.approve ── ✓ card.move() ── ⟩ report.build ── ⚙ prompt.rewrite ── ░▒▓ ▓▒░ ── > agent.idle ── │ review.approve ── ✓ card.move() ── ⟩ report.build ── ⚙ prompt.rewrite ── ░▒▓ ▓▒░ ── > agent.idle ── │ review.approve ── ✓ card.move() ── ⟩ report.build ── ⚙ prompt.rewrite ── ░▒▓ ▓▒░ ── > agent.idle ── │ review.approve ── ✓ card.move() ── ⟩ report.build ── ⚙ prompt.rewrite ── ░▒▓ ▓▒░ ── > agent.idle ── │ review.approve ── ✓ card.move() ── ⟩ report.build ── ⚙ prompt.rewrite ── ░▒▓ ▓▒░ ── > agent.idle ── │ review.approve ── ✓ card.move() ── ⟩ report.build ── ⚙ prompt.rewrite ── ░▒▓ ▓▒░ ── > agent.idle ──

⟩ knowledge.grow ── █ version.tag ── ├── test.validate ── │ loop.continue ── ██▓▒░░▒▓ ── $ agent.wake ── ⟩ knowledge.grow ── █ version.tag ── ├── test.validate ── │ loop.continue ── ██▓▒░░▒▓ ── $ agent.wake ── ⟩ knowledge.grow ── █ version.tag ── ├── test.validate ── │ loop.continue ── ██▓▒░░▒▓ ── $ agent.wake ── ⟩ knowledge.grow ── █ version.tag ── ├── test.validate ── │ loop.continue ── ██▓▒░░▒▓ ── $ agent.wake ── ⟩ knowledge.grow ── █ version.tag ── ├── test.validate ── │ loop.continue ── ██▓▒░░▒▓ ── $ agent.wake ── ⟩ knowledge.grow ── █ version.tag ── ├── test.validate ── │ loop.continue ── ██▓▒░░▒▓ ── $ agent.wake ──

$ tool.register ── ⚡ auth.oauth() ── │ ├── data.fetch │ └── parse.json ── > _ complete ── ░▒▓██▓▒░ ── $ tool.register ── ⚡ auth.oauth() ── │ ├── data.fetch │ └── parse.json ── > _ complete ── ░▒▓██▓▒░ ── $ tool.register ── ⚡ auth.oauth() ── │ ├── data.fetch │ └── parse.json ── > _ complete ── ░▒▓██▓▒░ ── $ tool.register ── ⚡ auth.oauth() ── │ ├── data.fetch │ └── parse.json ── > _ complete ── ░▒▓██▓▒░ ── $ tool.register ── ⚡ auth.oauth() ── │ ├── data.fetch │ └── parse.json ── > _ complete ── ░▒▓██▓▒░ ── $ tool.register ── ⚡ auth.oauth() ── │ ├── data.fetch │ └── parse.json ── > _ complete ── ░▒▓██▓▒░ ──

│ dispatch.send() ── ✓ task.close ── ⟩ metric.track() ── ⚙ agent.reflect ── ▓▒░ ██ ░▒▓ ── > loop.end ── │ dispatch.send() ── ✓ task.close ── ⟩ metric.track() ── ⚙ agent.reflect ── ▓▒░ ██ ░▒▓ ── > loop.end ── │ dispatch.send() ── ✓ task.close ── ⟩ metric.track() ── ⚙ agent.reflect ── ▓▒░ ██ ░▒▓ ── > loop.end ── │ dispatch.send() ── ✓ task.close ── ⟩ metric.track() ── ⚙ agent.reflect ── ▓▒░ ██ ░▒▓ ── > loop.end ── │ dispatch.send() ── ✓ task.close ── ⟩ metric.track() ── ⚙ agent.reflect ── ▓▒░ ██ ░▒▓ ── > loop.end ── │ dispatch.send() ── ✓ task.close ── ⟩ metric.track() ── ⚙ agent.reflect ── ▓▒░ ██ ░▒▓ ── > loop.end ──

Bring us work that needs its own loop.

We’ll map the actions, feedback, and source of truth with you.

Book a discovery call