Bailey
Product Manager · proposes what is next
Reads the backlog you attach in Jira or Linear and proposes the next item, with the evidence attached.
neverships. A proposal always waits for a person.
Two ways in · one shared record · one rate card
Agent teams that carry the work.A coding agent that costs 60% less.
Both run on the same vetted, US-hosted open weight models, write to the same decision ledger, and bill the same meter. Start with whichever problem is louder this quarter. The other one is already here when you need it.
Agent Teams
Run agent Teams
Six named Workers carry a ticket from the backlog to a merged pull request, managed like your engineering team: a plan you approve, a review you can read, autonomy earned on the record.
If the work is louder ↓YAGNI Code
Cut the coding-agent bill
A drop-in desktop and CLI coding agent, or route Claude Code and Codex through YAGNI with one command. 60% or more under frontier API rates, with the cost printed as you go.
If the bill is louder ↓30 minutes with the founder · bring a repo and a backlog, or your usage export
If the work is louder
Agent Teams, managed like your engineering team.
Six named Workers carry a ticket from the backlog to a merged pull request. You approve the plan, read the review, and loosen the gates as the record earns it.
01 · Meet the Workers
Six crafts. One Team.
Each Worker has a job, a boundary it never crosses, and a place on the record. Every one of them starts at Training.
Reeve
Staff Engineer · critiques the plan
Critiques the plan before a line is written, against the rules you set, on the strongest reasoning model.
neverauthors the plan and never blocks it. Speaks only on consequential changes.
Wright
Builder · builds to a draft pr
Takes an approved plan to a pull request in a cloud sandbox and answers review rounds until the reviewer is satisfied.
nevermerges. The pull request is the finish line.
Proctor
Reviewer · reviews every pr
Reads every pull request through its checks (deep review, business fit, database, security, tests) and posts one review.
neverblocks a merge. Approves only once the Engagement has earned it.
Fletcher beta
QA · tests it in a browser
Boots the app in a sandbox, writes a test plan from the change, walks it in a browser, and attaches the video.
neverchanges application code and never weakens a test.
Harper
Chronicler · reports what happened
Composes the Brief and the Slack standup: what shipped, what is in flight, what needs a person.
neveracts. Reports only.
02 · One ticket, start to finish
Watch one work item cross the Team.
A ticket comes off your backlog and moves proposal, plan, build, review, test, report. Every step is credited to the Worker that did it, every gate waits for a person, and the whole thing is on the record when it is done.
Frontend Team · NMT-214 · add a geolocation providerITEM COST $1.84
- baileyproposed from the Jira backlog · NMT-214 · add a geolocation provider · evidence attached
- slackPM tagged in #eng-frontend · business rule added: fall back to postal code when GPS is denied
- youproposal accepted · start planning
- wrightplan drafted in a sandbox · 3 questions for you, 2 answered from the decision ledger
- reevecritique: index the lookup table · wording tightened · no other concerns
- youplan approved · one edit
- wrightbuilt · 14 tests added · draft PR #482 opened
- proctorreview: 2 findings, security and db · both fixed by Wright · approved
- fletcherapp booted in a sandbox · 6-step test plan · video and 9 screenshots attached
- youresult accepted · merged
- harperstandup posted to #eng-frontend · NMT-214 shipped · receipts attached
- The Team waits here. The plan is yours to approve.
proposal · plan · build · review rounds · browser test · report · every gate answered by a person
Your gates, your Ladder
At Training every step waits for you. As the record fills, you promote an Engagement one rung and it stops asking about the things it has earned. Merging stays with a person at every rung.
The PM in the loop
Tag a product manager in Slack at the proposal, and their answer lands in the plan before a line is written. Domain knowledge goes in at the front, not in a review comment at the end.
03 · Manage them like your team
Teams that mirror your org. Autonomy that is earned.
A Team mirrors one of your engineering teams, never a repository, with its own repos, Playbook, budget, sandbox, and one accountable person. A Worker serves it through an Engagement: when it runs, what it may do alone, and what it knows here. Proctor can approve on the web app and only advise on the infrastructure repo.
Every Engagement starts at Training. Promotion is a button you press when the track record supports it, never a switch it flips itself, and merges and anything irreversible stay gated at every rung.
Training
Supervised
Autonomous
promoted per Engagement, on the record · the floor on irreversible acts never moves
Frontend Teamweb app · design system · 2 repos
reports to Sam- baileysupervised
- reevetraining
- wrighttraining
- proctorautonomous
- fletchertraining
- harpersupervised
Backend Teamclaims API · pricing · 4 repos
reports to Marcus- baileytraining
- reevesupervised
- wrighttraining
- proctorautonomous
- harpersupervised
two Teams · one shared decision ledger · each Engagement on its own rung
Frontend Team · Playbook
rules the Team has learned · read before every plan
Price a claim as the contracted rate times the modifier, minus the member's remaining deductible. Never estimate it.
Any migration that adds a lookup needs an index in the same PR.
Fall back to postal code when the browser denies location.
Propose at most three items each weekday morning.
plans · decisions · playbook · exportable to confluence or drive
04 · The system of context
The plan, the decision, and the rule live in one place.
Every plan, every answered question, and every decision behind a merge lands in one shared record that the next plan reads first. The Playbook is where judgment becomes rules: a domain expert writes the pricing formula in plain words, an engineer writes the migration guardrail, and when your decisions show a pattern the Team proposes the rule for a person to adopt or dismiss.
It is your knowledge, not ours. Export it to Confluence or Drive any time.
05 · Start with one gate
Already running your own factory? Add a gate.
Three Workers only check output and never change code. Engage one on one repository beside whatever you run today, at a fraction of a frontier-model reviewer's cost. The builders join when the record earns it.
Proctor
Reviewer
On one repo, next to the reviewer you already run. Two reviews on the same PR is fine: when they agree, you know, and when they differ, you learn something.
neverblocks a merge. Approves only once the Engagement has earned it.
Reeve
Staff Engineer
On plans, before the code exists. The cheapest review is the one that happens before the first line is written.
neverauthors the plan and never blocks it. Speaks only on consequential changes.
Fletcher beta
QA
On pull requests, in a browser. A test plan written from the change, walked in a sandbox, with the video attached to the PR.
neverchanges application code and never weakens a test.
If the bill is louder
YAGNI Code. The same models, 60% or more under frontier rates.
A drop-in desktop and CLI coding agent, or the Claude Code and Codex your engineers already run, routed through YAGNI with one command. Nothing else about the workflow changes.
YAGNI Code · desktop, CLI, or the harness you have
Seats are becoming API rates. Keep the workflow, cut the bill.
Frontier vendors are moving growing teams off flat subscriptions and onto usage pricing. At those rates a power user clears$9,000 a month and a typical developer runs near $4,000, for the work your team already does today.
Adoption is not the problem. When most of your PRs are agent-touched, the spend scales with your best habits. The bill is the problem, and it is the one thing here that does not need a new workflow to fix.
The open weight models are cheap and competitive. There is no question in my mind, this is where we are all going to land.
How 60% is possible
The best open weight models are now genuinely good at a fraction of frontier API prices. YAGNI serves them from US hosting under zero data retention, evals every release before it serves a request, and a router places each step on the cheapest model that holds quality. Routine work never pays top rates.
Drop-in, two ways
Run YAGNI Code as a desktop app or a CLI; your skills, plugins, and MCPs slot straight in and the muscle memory carries over. Or keep Claude Code or Codex and route it through YAGNI's models with one command: yagni connect claude-code.
The same record as the Teams
Every session reads and writes the decision ledger the agent Teams use, so an engineer's correction today is in tomorrow's plan. Usage shows spend per developer and per model, with the frontier benchmark alongside.
Pilot free with two or three developers beside the agent they use today, andread the comparison from your own sessions, not our deck.
06 · We meet you where you are
Either way, nobody is asked to move.
The web app
Front, Work, the case file, the Workers directory, Usage. Where an engineering lead runs the Teams and reads the record.
Claude Code or Codex
Connect the harness you already use to the YAGNI MCP server: pull the queue, ask Reeve for a critique or Proctor for a review, read Fletcher's evidence, publish a review as a recorded Decision.
Slack
Bailey tags the product manager at the proposal, Harper posts the standup, and a reply in the thread lands on the record.
07 · Run it like an admin
Built to pass your security review.
US-only inference · zero retention
Every model YAGNI serves runs on US-hosted providers under zero data retention agreements, and the routing never leaves that lane.
Isolated sandboxes
Each Team's builds and browser tests run in their own cloud sandbox. Workers reach repositories only through the GitHub App you install, never a personal token.
Your identity provider
Google single sign-on and Okta today, with per-developer and per-Worker usage visible to admins from day one.
SOC 2 Type II underway · BAA available
The monitoring window is in progress. The trust portal, a DPA, and a Business Associate Agreement are ready before your pilot starts.
08 · Pricing
One meter. Developers and Workers alike.
One rate card with four lanes, metered per token, the same for an engineer in YAGNI Code and a Worker on a Team, at 60% or more under comparable frontier API rates. No seats, no platform fee, humans free. Caps per Worker and per day keep it predictable, and the rates are shared on the call.
Thirty days, one repo, your backlog
Bring one repo and a backlog. Meet the Team that takes it.
On the call we connect a repository and a backlog, engage the first Workers at Training, and leave the first work item at your proposal gate. The record starts that day.
30 minutes with Jack, the founder · no migration project on the other side