Lexenne create

I never write a prompt.

The work I keep for myself is the ADR (architecture decision record): pressure-testing the design one-on-one with the architect agent until the shape of the work is right. The architect files the tickets from the finalized decision - scoped, prioritized, and auto-assigned.

The sign-offs I keep - ADR approval, promotion out of Intake - are a hand on the wheel by choice, not steps the machinery requires: the steward agent promotes what it files, the repair loop promotes its own, and the system runs the same when I step back. What the hand buys is priority steering across a hundred-plus-ticket backlog.

Hands-off loops are not one feature; they form from the state of play

The formal one: the evaluator finds a defect, files the ticket, and the repair loop promotes and completes it - no human touches it. Organic ones run alongside: the steward spots a backlog pattern, pulls the data, and messages the architect to write the ADR and file the tickets; simpler fixes are filed directly or routed to the debugger; the steward also files tickets to automate work it notices itself doing by hand. Lessons feed back into each role's working memory, so a lesson doesn't have to be learned twice. That's the shift that matters: removing the human from every step where a human was never the bottleneck.

create is architecturally different. The backlog is a live state machine with transactional guarantees. A ticket promoted to ready is picked up automatically: the system writes the agent-scoped prompt, launches the session, and dispatches by priority with no clipboard and no paste. Each agent registers its session, claims work, executes under a structured SOP, runs RED→GREEN→REFACTOR, and submits for review through a single HTTP API. Shared scratchpads carry context across sessions, ISC (Ideal State Criteria) tables make "done" binary, and structured reflection captures what the next agent needs.

One protocol, two products.

create is one half of a pair of products that share a single seam. create is where the work happens; remember is the memory that persists across it. They speak one protocol - SLF (Substrate · Lens · Frame) - over a single interface: create reads context and writes captures to remember through render(substrate, lens, frame) → receipt, and every access leaves a signed receipt you can audit. The two are being split into separate codebases whose only shared code is SLF.

The receipt-and-grant enforcement across that seam is designed; the memory it runs on is live today.

It makes you an architect of systems that develop.

When agents do the building, the scarce skill is no longer writing the code. What's left is judgment: defining what good looks like, evaluating what the agents produce, and designing the system that makes them reliable. That is the work that compounds.

Because the building no longer requires hand-writing the code, the door opens to anyone who can define what good looks like and hold the work to it. And it raises the value of deep technical knowledge rather than retiring it: shaping the incoming requirements, the scope, the priorities, and the definition of done is architect's work, and it is what separates a system that produces the right thing from one that produces plausible output quickly.

What create unlocks isn't "AI writes my code." It's institutional execution capacity: the ability to run more work, in parallel, with higher confidence than any one person could sustain, and to get measurably better at it over time, because remember compounds what the agent network learns.

The harness is the product.

create is not a collection of prompts and a kanban board. It is a purpose-built engineering system designed to make agent-first development reliable, auditable, and continuously improving. Every component - the session protocol, the approval queue, the work packages, the automated Launch / Relaunch dispatch loop, the TDD enforcement, the scratchpad continuity - exists to answer one question: what capability is missing, and how do we make it legible to the agent?