01 / 13
← → to navigate · E to edit
Edit mode — click any text · Ctrl/⌘+S to download
Braid · Seed 2026

Generation made
software abundant.
Braid makes it ownable.

The work layer for AI software.
human → agents → one governed change
Why now · 01
Monday morning. 37 agents finished overnight.

Software is becoming multiplayer.
The workflow still assumes single-player.

37agent branches
112CI runs
19repair attempts
8security flags
0shared memory of why
Braid
Why now · 02

Output scales with compute.
Absorption does not.

  • Producing work is cheap, abundant, parallel.
  • Absorbing it — review, verify, govern, trust — is capped by human judgment.
  • Every agent you add widens the gap.
absorption agent output review · CI · security · merge · release →
Braid
Why now · 03

Git organizes around the diff.
The next layer organizes around the work.

A diff shows what changed. It can't carry intent, rejected paths, evidence, approvals, cost, or risk — yet it's still the artifact teams decide on.
the diff
retry.go
+52 −3
intent evidence rejected paths approvals cost policy · risk
Braid
The pain · 04

The team can produce more work
than it can confidently absorb.

Reviewers
“Please reconstruct what this was trying to do.”
CI & Security
“Unknown change — run the broad checks.”
Leaders
“Can this land? On whose judgment?”
One opaque AI change is annoying. Dozens through review, CI, security, and release become an intake problem.
Braid
The pain · 05

Human judgment is the one part of the loop
that doesn't scale with compute.

People break first
You add agents faster than trusted reviewers. The same scarce people become the last line of coherence — judging more work, each on a partial view.
Then infrastructure
When systems can't tell a rename from a migration, they run every check on everything. Agent volume turns that into thousands of runs.
Most teams haven't hit this wall yet — but every team that keeps adding agents is walking toward it.
Braid
The pain · 06

The root pain is neither volume nor context loss.
It's ungoverned work at the point of change.

AI work reaches the moment of change with no durable way to prove what happened, who approved it, or whether it should land — and no way to enforce one policy across agents that don't share tools, memory, or authority.
First user
The developer / tech lead getting agent work across a boundary.
Buyer
The eng / platform / DevEx / security leader who needs more good work to land with confidence.
Braid
What we're building · 07

Braid is the
work layer
for AI software.

It captures the work behind a change, makes it usable to people and agents, and gives teams a control point for what becomes real software. Not “AI wrote these lines.” The unit is the work.
work package
↳ intent · the goal behind it
↳ scope & contributors
↳ approaches tried & rejected
↳ verification & cost
↳ approvals & policy
the diff +52 −3 — no longer alone
Braid
What we're building · 08

Capture is the commodity.
Multiplayer is the moat.

CodexClaude CodeCursorCopilotGitHubGitLabterminalCI CLI · API · MCP
  • A free, open-source developer product — works where you already are.
  • Provider-neutral. A signed record that travels with the code — and survives even if Braid's servers don't.
  • Useful before anyone buys. That ubiquity is the land-grab; connected multiplayer control is the business.
Braid
What we're building · 09
Live demo

A 40-minute reconstruction
becomes a 4-minute decision.

Without Braid
$ git blame retry.go:42 a1b2c3d (claude-code-agent) return Retry(cfg, fn)
Three weeks later it fails in prod. Why this backoff? What was rejected? You don't know. You guess.
With Braid
$ git log --show-notes=braid HEAD Agent: claude-code · read: retry.go, CLAUDE.md, #138 Considered 3 approaches: ✗ external lib — CLAUDE.md: prefer stdlib <40 lines ✗ hardcoded — #138: thundering-herd risk ✓ 35-line helper with jitter Reviewer: alice (vouched)
The reasoning that produced the commit is on the commit. The answer is a lookup, not a guess.
Braid
What we're building · 10

The decision no longer
depends on the diff alone.

  • The work sits beside the diff: intent, runs, cost, verification, approvals, Gate status.
  • Many human + agent branches fan into one reviewed change.
  • The PR stays familiar. The decision changes.
human agent agent agent one reviewed change
Braid
What we're building · 11

The same checks teams already run —
now fired in proportion to the work.

Mechanical rename→ light path. Low-risk work merges fast.
Thin evidence→ deeper verification, where it's actually warranted.
Behavioral migration→ scarce human judgment, where it matters.
Because the work carries intent & scope, the Gate governs proportionally — and enforces one policy across every agent. We don't run your CI; we give it the trusted intent to run less of it.
Braid
What we're building · 12

Own the moment
AI work becomes
real software.

  • Free + open-source substrate spreads.
  • Paid team workflows coordinate & govern multiplayer work.
  • The PR is the beachhead. The company is the control layer for the AI-native SDLC.
review · CI · security · release · incidents · planning
Braid — the work layer
code · diff · Git
Braid · braidkit.io