AI Factory
AI across the software development lifecycle
AI already helps write code, but preparing tasks, checking results and starting each next step still demand attention from the team.
Discuss this workshopAn example exercise
Illustrative task. We choose the actual exercise around your team’s work and tools.
The task: add search by name to an internal list. Clarify the request, implement and verify the change, then release to a test environment after approval. Try repair and rollback with deliberately faulty behaviour too.
From the request to running software.
Explore the whole workflow through one small task. The practical release uses a test environment. Open each step to see the different roles of AI and the team.
-
Clarify what is needed
AI helps gather context and open questions. Define the expected result and how it will be verified.
People approve the goal and scope.
-
Plan the change
Find the relevant code and documentation. Define available tools and permitted changes, then review the plan.
Access and permitted actions are defined in advance.
-
Implement and run checks
The agent works in an isolated environment, runs tests, revises from feedback and collects the verification results.
Repeated failures or missing decisions trigger a stop and a request for help.
-
The team reviews the result
The task, change and check results are available together. Feedback can start another revision round.
Approval belongs to people; passing tests alone are not approval.
-
Release and verify the behaviour
The approved change reaches a test environment through the release process. Check the affected behaviour there too.
Verify the released version actually works.
-
Turn a failure into repair or rollback
Assess behaviour using logs and measurements. A prepared failure demonstrates stopping, repairing and rolling back.
Agree when intervention is needed and who decides.
During the exercise, look at elapsed time, manual interventions, review effort and run costs. Use these to decide where further automation is worthwhile.
How we work
-
Design the task’s path, access boundaries and human decisions.
-
Connect AI work with checks and approval.
-
Try a failure case and define the steps for your own adoption.
What stays with your team.
- A working example from task intake through post-release checks.
- Explicit permissions, approval points and failure-handling rules.
- An adoption plan with measurement and operational considerations.
Making time for the work
A multi-session programme with building and trials. Design, implementation and failure cases each get time. A complete production rollout is a separate engagement.
We agree session count and length, group size, preparation, follow-up and fees before starting. Environment preparation and practice get their own time allocation.
What do we need to start?
Practical AI experience, a small development task, runnable tests and a test environment. A prepared example is available; a Berek subscription is not required.
Explore the foundation programmeBuilding on the foundations
This workshop works across the full workflow, so your team needs to know how to give AI a task and verify its output. The foundation programme develops that practice. If your team already has it, you can start directly with AI Factory.
Explore the foundation programmeThe method is yours. Berek can provide the foundation.
We demonstrate the workflow using Berek. You can implement the method with your own tools; if you prefer an existing foundation, we can also help introduce Berek. The workshop stands alone, and adopting Berek is a separate decision.
Explore Berek