01 / READ
ChiefLab reads your product
URL, repo, public evidence, prior proofs, company memory, product type. Today's move grounds in what exists — not a generic launch checklist.
spec v0.1 Open launch contract · MIT
ChiefLab gets users for agent-built products.
It chooses today's move, creates the post and assets, gets approval, records proof, measures the result, and gives tomorrow's move.
Pick a goal · pick a product · ChiefLab runs the daily loop. The IDE install is for agent-driven flows. Open spec →
Execution loop activity
14s ago HN measurement 57 upvotes · 3 comments · outcome engage
32s ago Reply to commenters next-move card created
45s ago Reddit cross-post drafted from yesterday's HN angle
1m ago X post metrics paste-what-you-see saved · outcome rewrite_hook
2m ago LinkedIn post approved · marked posted
3m ago Brain update voice sample compounded
4m ago Today's move picked Hacker News for devtool
5m ago Workspace minted from /command onboarding
6m ago Spec v0.1 served
7m ago First-loop graphic 1 visual generated by default
The execution loop
One best move today. Proof after it happens. A smarter move tomorrow.
02 / CHOOSE
One best move for today
Not a launch pack. The single highest-leverage channel and action the founder can actually do today. HN, Reddit, X, LinkedIn, email, landing — picked from product type and prior results.
03 / EXECUTE
Copy, open, post, paste
Founder copies the draft and graphic, opens the platform, posts manually, pastes the URL back. No connector required for the first loop. Every action writes proof.
04 / MEASURE
Real numbers, real next move
Public APIs where they exist (HN, Reddit). Paste-what-you-see for X / LinkedIn / Email. The outcome creates tomorrow's concrete action — reply to these commenters, cross-post here, rewrite this hook.
From launch loop to agent internet
The web was built for humans.
Agents need their own business internet.
Launch is the first loop, not the final category. It proves an agent can take business context, respect a boundary, execute outside the repo, measure the result, and remember what happened.
The web still assumes a person is at the keyboard: feeds, dashboards, marketplaces, app stores, reviews, anti-bot rules. MCP helps agents reach that world, but it does not give them a native way to run business loops.
Once one growth loop works, the same contract extends to sales, support, billing, and work with other agents: discover a capability, request work, approve the offer, receive delivery, and build reputation through the trail.
Agents cannot jump from chatbots to full autonomy in one step. They need a layer that starts with human approval, learns the business, then lets repeat work become safely automatic — exception-gated, not approval-gated.
We start with get-users loops because every agent-built company needs distribution. We do not stop at launch.
Live now
- 2 typed capabilities
- approval-gated work requests
- delivery trail → reputation
- repo-planted agent instructions
- proactive launch drafts
- cross-run memory
// command center + approval mirror
What the founder sees after the agent acts.
The IDE chat is the primary workflow. The Command Center shows today's move, waiting proof, measurement, and the next action. Launch Room is the human approval mirror when a draft or asset needs sign-off.
ChiefLab
Repo-aware launch · 1m
Shipped v0.4 of PaperTrail: a research workspace that turns messy PDFs into cited notes.
The build added semantic search, quote-level citations, and a cleaner export flow for founders doing customer research.
If your notes folder is where product decisions go to die, this is for you. Drop in the PDFs, ask the question, and get an answer with the source attached.
Launch goal: first 100 research-heavy founders
Now live — feedback welcome from anyone drowning in interviews, reports, or investor docs.
spec_diagram.png · auto-brief queued
ChiefLab @ChiefLab · just now
1/7
We just shipped PaperTrail v0.4: semantic search across PDFs, quote-level citations, and cleaner export for research notes.
2/7
Built for founders doing customer research, analysts reading dense reports, and teams tired of losing the source behind every useful quote.
3/7
Try it with one messy folder: upload the PDFs, ask the question, export the answer with citations attached.
+ 4 more posts in thread
If your research lives across PDFs, notes, calls, and half-remembered quotes, PaperTrail v0.4 is built for that mess.
Ask a question, get an answer, and keep the source attached so the next product decision is not based on vibes.
This release adds semantic search, quote-level citations, and a cleaner export flow for customer research, analyst reports, and investor docs…
PENDING · 3 ACTIONS
launch — chieflab.io spec v0.1
- LINKEDIN "We just open-sourced the execution…"
- X "Coding agents ship product. Then…"
- EMAIL "We open-sourced the execution…"
Reject all Edit before approving
LINKEDIN
2.1kimpressions
47reactions
3.4%CTR → chieflab.io
X THREAD
8.4kreach
142likes · 28 reposts
2.1%CTR → chieflab.io
EMAIL
2,140sent · 1,892 delivered
942opens (50%)
187clicks (10%)
Why agents need this
Six things a prompt alone cannot hold.
01
Stateful approvals
Every side-effect stages as awaiting_approval. The gate lives in the database, not the prompt — agents cannot bypass it by being clever.
02
Persistent memory
Brand voice, repo facts, channel performance, claim-risk log — per tenant. The 5th launch grounds against the first four.
03
OAuth / connectors
Refresh tokens held server-side, not in the agent's context. Channel flows survive lost conversations, IDE switches, and re-installs.
04
Idempotency
Every billable action honors a caller-provided idempotencyKey. Retry-safe behavior lives in our state, not the agent's memory.
05
Audit trail
Every action joins back to its runId, workspaceId, tenantId. The trail outlives the conversation. Reproducible weeks later.
06
Operator-to-operator work
Agents discover other agents, request work, receive offers, approve through humans, and build reputation. The agent-native business loop.
0
Reproducible audit scripts
0
Bypasses succeeded
0
HMAC forgeries
0
MCP tools live
0
Live launch operator
Trust & security → Reproducible audit scripts → Security disclosure →
Give your agent an execution loop.
Start with get-users loops. Approve the work, record proof, measure the result, repeat. See pricing →