202 Tasks Completed Ahead of Go-Live
A real estate certification company in France faced a growing list of changes to its system. With the AI Task Line, AI agents took over writing the code, while the Doodex team oversaw every result through testing and human validation.
Running a Complex Certification System Without Losing Control
Before working with Doodex, the company faced several operational challenges:
In France, certifying a building can take years, from construction to final certification. With complex rules that keep changing and a government body watching, every document and every piece of information has to be handled correctly.
Information was hard to find
The team spent days looking for information that already existed in the company.
The same documents were requested again and again
Customers had to send the same papers more than once.
Deadlines were missed
Information was often requested too late, so certifications were delayed or even failed.
Company knowledge was at risk
Proof and key information could leave with employees, and access rights were not under control.
Since 2023, Doodex has worked with the client to rebuild their business system. The old CRM was retired, and the client is now moving off Salesforce to the new system.
In autumn 2026, the project entered its next phase: testing daily operations and certification file management ahead of a go-live planned around 7 December 2026.
The number of changes to deliver kept growing. The team needed a way of working that could handle more tasks without giving up quality or oversight.
AI Agents Write the Code, the Doodex Team Checks Every Result
On 23 October 2026, Doodex started running the work on this project through the AI Task Line. Want to know what else we do? See Doodex services.
One line, from request to report
Instead of handling requests as large pieces of work that are hard to track, every request is split into small tasks with a clear goal, acceptance criteria and tests.
- Every request becomes measurable tasks
- AI agents write the code
- Every task goes through two rounds of testing
- People stay in control
- The client gets clear reports
Each task has a goal, acceptance criteria and the tests it must pass.
Every request becomes measurable tasks
Each task has a goal, acceptance criteria and the tests it must pass.
AI agents write the code
All code in this scope of work has been written by AI agents since 23 October.
Every task goes through two rounds of testing
First on the developer's copy of the client's system, then on a shared test server.
People stay in control
A developer or the project manager validates by hand when a task keeps failing or needs a human decision.
The client gets clear reports
Every finished task comes with a short report in French.
The AI Task Line Flow screen
Eight stations: six on the developer's computer, two on a shared test server.
Click any station for an explanation, or run the simulation.
With this flow, AI agents handle the coding, while developers, testers and the project manager focus on oversight, testing and the decisions that need human judgement.
“With one shared source of truth and complete task descriptions, the AI's work becomes precise and accurate. Code gets written faster and review runs automatically. Our developers, testers and I now focus on guiding and understanding every task we give the AI agents, including when a meeting decision cancels a ticket. Human validation is still essential for the decisions AI can't make on its own.”
Project Performance Summary
The project started on 14 October 2026, with AI Task Line taking over on 23 October. By 18 November, the Doodex team had completed 202 tasks through a structured build-and-test workflow.
Tasks delivered
Median from task created to done
Tasks created
Tasks delivered per week (full weeks)
Tasks delivered in the busiest full week
Of code written by AI agents since 23 October
More Than a Task Count
Every result still goes through testing and oversight before it reaches the client. The per-task reports in French also got positive feedback. The client's team needed time to get used to finding the reports in a new place, and asked for a shorter format.
Heading to Go-Live
Production go-live is planned around 7 December 2026, after the client's IT team returns from leave.
Measurement Method
All figures come from the Doodex project tracker, as of 18 November 2026. A task counts as done the first time it reaches the Done stage. Doodex did not record delivery speed the same way before using the AI Task Line, so this case study makes no claim of a percentage speed gain over the previous way of working.
Four Tasks From the AI Task Line
Real client requests, rewritten in plain words. Time runs from when the task was created to when it was first done.
Hide the edit (pen) icon on fields the user is not allowed to change.
The pen now shows only where the user can really edit, on every screen of the internal platform.
Let every user edit someone else's activity, but not delete it.
Any user can change an activity's date, notes and assignee; deleting stays blocked. After a due-date change, the activity stays in the planned list.
Tell apart an e-mail still to send from an e-mail already sent.
Two different icons on every screen: activity list, file history, timeline and home page.
Use the same contact and address fields on every online form.
All forms share the same layout. Company details show only when the applicant is a company, and the postcode sits between the street and the city.
Our Team Does the Work, Helped by the Machine
Since the AI Task Line started on 23 October 2026, the number of tasks the whole Doodex team completes each week rose from an average of 36 to 127–285.
tasks a week on average, January to mid-October
tasks a week in the 4 full weeks since 23 October
tasks done in those 4 weeks
Source: Doodex hub, all projects, first move to a Done stage per task. Mass clean-ups removed. Last bar = 16–18 Nov. The weeks of 21 Sep to 12 Oct were already above average before the AI Task Line.
FAQ
The questions buyers ask us most about the AI Task Line.
Will the AI Task Line replace programmers?
No. On this project, AI agents write and test the code, while developers stay responsible for preparing tasks, checking results and stepping in when a task keeps failing.
Is AI-generated code safe to use?
Every task goes through two rounds of testing before the result reaches the client. Tasks that keep failing are also checked by hand. Validation stays a core part of every piece of work.
How long does the AI Task Line take to finish one task?
In this case study, the median time from task created to done was 3.1 days. That figure comes from one project and is not a guaranteed turnaround or SLA for every project.
Sources: Doodex project tracker and task list, 23 Oct – 18 Nov 2026 · AI Task Line guide v0.11.0, 9 Nov 2026 · Client background: Doodex migration report and CEO interview.
From consultation to support after go-live
Doodex helps companies design, build and manage business systems that fit their operational needs. Our services cover ERP implementation, system integration, custom solution development, data migration and support after go-live. With the AI Task Line, our team works more efficiently, with human validation at every step.