00:00
We save the output.Trace saves the path.
AI now does real work across chats, browsers, editors and terminals. Trace preserves the path that produced it, so the work can be learned from, repeated and audited. Starting with AI-native work.
AI is making work easier to produce and harder to understand.
- then
- promptresponse
- now
- intentplantool callsbrowsingeditscommandsdecisionsoutput
We have chat history.We have version history.We still don’t have work history.
Work is getting more autonomous and more fragmented at the same time. We get more output, and lose the causal chain between what we wanted, what the AI did, what we changed and what finally worked.
Keep the path, and work becomes something you can learn from, repeat and audit.
01
Learn
As AI does more of the execution, expertise moves into the judgment: what to ask, what to inspect, what to reject, what to change by hand.2
- today
- the code
- trace keeps
- the reasoning that shaped it
- for
- workshops & cohorts · tutorials · onboarding
02
Repeat
A workflow that worked last Tuesday shouldn’t have to be rediscovered this Tuesday. You can’t automate a process you never captured.
- today
- files, prompts, outputs
- trace keeps
- the procedure
- for
- engineering teams · PR handoffs · ops workflows
03
Audit
As agents do consequential work, proof of process matters almost as much as proof of output: what it saw, what it did, what a human approved.
- today
- the result
- trace keeps
- the provenance
- for
- hiring · research & publishing · regulated work
Replayed by learners, teammates, reviewers, auditors — and future you.
Not a recording.A trace.
A trace is a structured record of a work session: the prompts asked, pages read, files changed, commands run and what came out the other end.
Captured across your tools → structured into one timeline → replayable by anyone you share it with.
searchable · replayable · shareable · redacted
Add Supabase auth to a Next.js app
“Design an auth flow for a Next.js app on Supabase. Magic-link sign-in, and /dashboard should be private.”
→ Suggested session refresh in middleware, RLS on profiles, server-side signIn()
supabase.com/docs/guides/auth/server-side/nextjs
supabase.com/docs/guides/database/postgres/row-level-security
2 pages · 6 min
- // TODO: auth
+ export async function signIn(email: string) {
+ const supabase = await createClient()
+ return supabase.auth.signInWithOtp({ email })
+ }
$ supabase db push
Applying migration 20260926_profiles_rls.sql…
Finished supabase db push.
$ vercel --prod
✓ Production ready [1m 48s]
Signed up with a magic link → landed on /dashboard
Outcome linked to the 5 steps that produced it
Software got too complex to change without version history.AI-native work is getting too complex to understand without process history.
Four things changed at once.
First, Trace records. Later, it may help you do it again.
From copilot to agent
Agents now plan, browse, run commands and edit files, often in the background. 90% of professional developers use AI coding agents at least weekly; 68% daily.3
Work crosses app boundaries
One session spans chat, browser, IDE, terminal and SaaS tools. No single app holds the whole story.1
Execution is cheap; judgment isn’t
People make most of the planning decisions; AI makes most of the execution decisions. The decision trail is the valuable part.2
Agents need accountability
Platforms are shipping audit logs for agent activity. “How did the AI arrive here?” is becoming a question you have to answer.4
Built for trust, not surveillance.
- Selective capture
- Record what matters, not everything.
- Private by default
- Nothing leaves your machine unless you share it.
- Redaction-aware
- Secrets, keys and personal data are caught before sharing.
- Human-readable
- Summaries you can read, not raw logs.
If AI is part of how you work, your process deserves memory.
Trace is being built in public, starting with people who build across AI chats, browsers and code.