It is not the model. It is the process.
The process decides success: repeatable, the same for every team member and steered by people.
- 1FetchReads the work and checks it is ready.
- 2RefineAsks your questions, writes the specification.
- 3ImplementBuilds the change.
- 4VerifyRuns the user flows and the tests.
- 5ReviseFixes what Verify found, then verifies again. Back to Verify. back to Verify
- 6MergeVersion, release notes, release.
It starts with your architecture.
Written down once, structured and machine-readable, in your own repository.
Every step gets its own context.
The agents work from what is written down; the architecture sets the context.
It moves with every change.
The documentation is updated after every change.
- your-application/
- architecture/
- decisions/Why it is built this way
- components.mdWhat each part does
- requirements/
- group-totals.mdREQ-1 to REQ-33
- test-cases/Linked to the requirements
- skills/Your own skills, shared and measured
- src/The code

Your processes, not ours.
One team runs several processes. Each one is built to fit the work, and your team can change it.
Development
A larger planned change.FetchRefineImplementVerifyReviseMergeScope change
Opens when building shows the approved specification no longer fits.What changed · your decisionRe-planBack into the pipelineYour own flows
A hotfix, an incident, a migration: shorter or longer, with their own gates.Your phasesYour stepsYour gatesYour skills
How it runs.
The Conductor runs the process step by step. The Watchtower brings every question to the right person, and their decision back.
Decides what is built, answers the questions, approves what goes live.
Sees every running change. When an agent needs a person, it asks the right one, with the evidence.
Runs your process: each step in order, each step handed to an agent with the right skill and the right context, on your own machines.
- Your repository and code
- Your tests and checks
- Your backlog
- Your own agents and skills
- Our skills, ready to use
- Your own AI subscription
- Product owners
- Architects
- Developers
Measured, then improved.
Each run is recorded step by step. What the numbers show is fixed in the process, so the next run is better than the last.
- 1
Run
The same process for each change, step by step.
- 2
Measure
Time, tokens, checks and rework, per step.
- 3
Analyse
Where does time go, which errors come back? How can we save?
- 4
Improve the process
A skill, a check or a step changes, not just one change.
and run again, better
- 9Aug
- 24Sep
- 30Oct
- Implement38%
- Verify31%
- Revise26%
- Refine4%
- Fetch1%
- Merge1%
A quarter of the time goes to Revise. That is the rework the next improvement goes after.
Your code stays yours.
The work runs on your own machines or cloud environment, with your own AI. Only permitted process data reaches TackForward.
Where the work is done
- Your code and repositoryNever leaves your environment.
- Your AIYour own subscription, such as Claude or ChatGPT.
- The runnerRuns each step with your skills or ours.
Where your team decides
- Every change in one viewFrom request to live.
- Your team's decisionsApprove, adjust or stop.
- Measures per changeTime, cost, checks and rework.
Six rules we work by.
- 1
Transparency builds trust
Every step and decision can be traced.
- 2
Responsibility is human
People set the direction and approve the result.
- 3
What gets measured gets done
Capacity, cost, quality and rework are measured.
- 4
Determinism beats reasoning
Fixed checks in software; AI for analysis and building.
- 5
Improve the process, not the product
Recurring errors are solved in the way of working.
- 6
A team effort
Product owners, architects and developers shape the method together.
Your first step takes two hours.
Bring one real challenge. In the free workshop we map where AI can strengthen your delivery.
A person from our team replies by the next business day.