For software teams working with AI
AI can make each specialist faster. Keeping the whole team aligned becomes the bottleneck.
Specialists and their agents can produce more within each role without improving the flow between roles. The wider team still has to interpret and review what moves between product, architecture, engineering, testing, and acceptance. Loft gives people and agents one human-ratified specification of what the product should do, and Yard puts every question, change and plan in front of the people who decide, and keeps it on the record.
Local acceleration can slow the whole delivery system.
Adding agents to each existing role speeds up work inside that role. It does not automatically improve the transition from product design to architecture, implementation, testing, and acceptance. More local output can create a larger queue at the next human gate.
Each role builds private context
A product manager's agents, coding agents, and testing agents can each work from a different interpretation of the same product. Their transcripts and local instructions do not create shared organizational intent.
Handoffs carry output without enough intent
The next specialist receives tickets, designs, code, or test results, then has to reconstruct the decisions, constraints, and acceptance basis behind them.
Human gates absorb the extra volume
Faster specialist-and-agent loops send more work into review and approval points that still depend on senior people rebuilding context and resolving ambiguity.
Read the full argument: When Every Role Gets Faster, Why Doesn't Software Delivery?
How Loft works
Give every role and agent the same product contract.
The contract can connect capabilities, requirements, components, constraints, quality targets, architecture decisions, vocabulary, acceptance criteria, and evidence for the selected scope. Specialists add work from their own discipline instead of creating separate versions of the product. Each change comes to the people who own it on the Yard page before it is committed, coding agents build from what they approved, and the result is checked against that same version.
People designated by the customer decide and ratify product intent. Loft records the accepted state, proposed changes, decisions and linked evidence, and Yard records how each change was made.
See how Loft worksSpecialists keep their expertise and share the same contract.
Product and domain specialists
Define outcomes, business rules, vocabulary, priorities, and acceptance conditions, and answer the agent's questions on the Yard page. Their agents help develop the product intent without turning a private workspace into the organizational contract.
Architects and engineers
Add technical constraints and decisions, approve the agent's build plan, then use coding agents to implement the approved scope. Their work stays connected to the intent it is meant to realize.
Testing and assurance specialists
Define independent scenarios, map executable evidence, and preserve failures or gaps instead of treating any passing test as complete proof.
Authorized product owners
Ratify changes to the contract on the Yard page and decide, with the record of how the work was done, whether the evidence is sufficient for acceptance.
Take one product contract through implementation, evidence, and a human decision.
Author
Say what the product should do. Your agent asks rather than guesses and writes the intent, requirements, constraints, decisions and acceptance criteria for the bounded change.
Ratify
Customer-designated people approve each change to the contract on the Yard page before it is committed.
Implement
The coding agent plans the build from the approved criteria. Once the plan is approved, it builds test-first, live on the Yard page.
Reconcile
Each test is linked to the criterion it verifies. Reconciliation shows what the evidence supports, what failed, what is missing, and what it cannot settle.
Decide
An authorized person reviews the result, the evidence and the branch's Yard Log, then accepts, rejects, or reopens the work.
A real run, on a fictional product
What the loop leaves behind.
-
RQ-9 If the account is overdue, then the service shall reject the pause request.
- AC-36 A pause request from an
overdueaccount returnsaccount_overdue. - AC-37 A pause request made after the overdue charge is paid returns
scheduled. - AC-38 A pause request from an account whose oldest unpaid charge is 13 days old returns
scheduled.
The approved contract: a requirement and the criteria that define done.
- AC-36 A pause request from an
-
Governing publish sequence 4
loft-website-exampleRQ-3RQ-8RQ-9RQ-10RQ-11RQ-12RQ-13RQ-14RQ-15RQ-16
The work scope: the pinned state and the requirements the agent must satisfy.
-
Candidate 1 ·
a7244b1Failed AC-36 A pause request from an
overdueaccount returnsaccount_overdue.Reopened
The evidence and the decision: a failure the agent's own tests missed, sent back before acceptance.
Latest product updates
-
Loft 2.3: Yard is available
Yard is where you and your coding agents meet while they work. It keeps a log on every branch of what they planned, asked, decided, and did, and a page on your machine shows all of it live. It ships with open-source Loft 2.3, in the plugin for Claude Code and Codex.
-
Loft 2.2: A status pane, a term list, and test links that notice rewrites
Loft 2.2 shows where a specification stands while a coding agent works on it, lets a specification name its concepts in a term list, and reports a test rewritten after its link to an acceptance criterion was reviewed.
-
Loft 2.1: Import OpenSpec projects and render them back
Loft 2.1 converts the main specs of an OpenSpec project into a Loft specification and writes a Loft specification back out as OpenSpec specs. A team using OpenSpec can start from the specs it already has and keep an OpenSpec copy while it adopts Loft.
Start with a new product or an existing system.
New product
Establish the intent, constraints, decisions, and acceptance criteria for one product area before or alongside implementation. Update the contract through reviewed changes as the team learns.
Existing system
Treat code, tests, documents, runtime observations, and expert knowledge as evidence. Resolve conflicts and ratify the intended behavior for one area before people or agents make a consequential change.
A practical place to start
Choose one product area with a result people can judge.
A first engagement takes one consequential capability, workflow, service, API surface, or change through the complete Loft loop, from the first draft of the contract to a human acceptance decision.
Open source
Try the Loft loop in your own repository.
The local edition is free and open source under Apache-2.0. Add the plugin to Claude Code or Codex and ask your agent to set up Loft: the specification, and each change's Yard Log, live in your Git repository, with no server. Use it on your own, then bring your team to the same specification.
Explore the open-source editionFor delivery organizations
Put customer authority into the delivery method.
Loft is exploring work with software delivery firms, consultancies, systems integrators, and implementation teams. Loft supplies the reusable product layer, including the Loft loop. A delivery organization brings customer context and implementation capacity. The customer retains domain authority and makes the final acceptance decision.
Learn about the partner pathLonger-term direction
Connect product decisions, software production, and operational evidence.
Current Loft already connects an approved product contract to bounded implementation and reconciliation, and Yard Client, the first part of the Yard pillar, records how each change was planned and decided. The Software Plant vision extends the same loop with Yard and Berth for broader production coordination and operational feedback.
Read the visionBring one product area or workflow worth making explicit.
Share a high-level, non-confidential description of the product context and where intent, authority, review, or acceptance is getting difficult. We will reply and set up a first conversation.