Context isn't intent.

AI writes code. Humans write software.

Git for code. Igni for intent.

Agents make the code. Someone has to own the calls.

Every implementation is full of judgment: tradeoffs, constraints, patterns, exceptions.

So teams add more context, write more memory files, add more review, and let agents plan and check each other. But the call still happens out of sight: inside a run, inside a memory file, buried in the code.

That is not directing software. It is finding out after the call has already been made.

Igni is the intent layer: where those calls stay connected to code, visible to agents, and shaped with humans as reality changes.

Not better context. Not blind trust. Infrastructure for intent.

The loop in motion

The moment a tradeoff becomes intent.

Agentic development does not need more places to watch work happen. It needs a place for judgment to become part of the work.

01

The work pushes back.

The work reveals what the plan missed — before it becomes a quiet choice in code.

02

Humans and agents shape the call.

Human judgment and agent reasoning meet on the same live surface — not a diff handed back hours later.

03

The call is now in force.

Not a note, transcript, or memory — graph-connected intent that future work has to build against.

This isn't hypothetical. In Anthropic's Fable 5 system card, the most common documented agent failure in their own engineering org is unverified judgment stated as fact (§2.3.3).

Want the full argument?

Read the essay →

What happens when an agent crosses a wall?

It surfaces as a finding, bound to the constraint it crossed — and becomes a call. It doesn't ship as a quiet choice.

What makes it work

One layer. Three jobs.

The intent layer has to connect judgment to code, let reality push back, and keep humans in the work while the call is still open.

Intent Graph

Where calls become force-bearing structure.

Decisions, patterns, requirements — typed and connected from strategy to code. Each relationship has force: governs, guides, depends on. Your agent can tell a wall from a preference.

Without it: intent is scattered context.

Intent Loop

The bridge between intent and code.

Agents build from graph-connected intent. What implementation reveals can change the call while work is live — and what crosses a constraint surfaces before it ships.

Without it: intent goes stale as reality changes.

Intent Stage

Where humans stay in the work.

Humans and agents share the same live surface while the call is still open. The agent's changes stay isolated until merge.

Without it: humans find out too late.

Day zero is not a blank graph.

Point Igni at your codebase. The graph builds itself from what's true — goals, features, implementations, patterns, constraints, mapped from real structure. You shape what should hold.

Day zero is curation, not data entry.

Charter access

For teams running agents on code that matters.

We're onboarding a small number of charter teams — engineering organizations shipping real production code through coding agents, ready to put their calls in force.

Read the essay →