WHAT'S NEW · 02 AI ASSISTANT
Improvements to the AI Assistant
In 2.8.0 the AI Assistant reads a failed run for you, proposes a fix, applies it, and re-runs the case so you see whether it now passes. Every change is yours to accept, and yours to undo. Until now there were two ways to deal with a failed run: work it out by hand, which is thorough but slow, or paste the log into a chat assistant, which is quick but blind to your run. This is both.
01 WHAT IT IS
What it does now.
The assistant sits beside your run details, or floats as a panel you can move and minimise. Each case keeps its own conversation, so you come back to the thread you left. It can change your project — case code, parameters, files, suites and schedules — and those changes appear in the Change Timeline, where you can undo them.
You will meet it in three places: Fix with AI on a failed run or a compile error, Edit with AI inside the editor with Ctrl/Cmd+K, and the chat itself.
Illustrative
02 WHY WE BUILT IT
Most of that work is gathering, not fixing.
Test maintenance is where automation quietly loses its value. Writing a suite is a project with a beginning and an end. Keeping it green as Maximo changes underneath it never finishes, and it lands on whoever opens the overnight results.
Doing that by hand is thorough, but it is slow and it needs someone who knows the case and the MXML. Handing the log to a general chat assistant is quick, but it only knows what you paste in, it cannot see the screenshot or the case, and it cannot change anything, so you are still the one doing the fix.
The assistant is given the facts, which case, which line, what the screenshot shows, what the log says, and it does the mechanical part. You keep the decision.
| By hand | A general chat assistant | 2.8.0: MaxTAF AI Assistant | |
|---|---|---|---|
| Sees the failure | Whatever you open: log, line, screenshot, video | Only what you paste in | Status, logs, failing line, screenshot and video, together |
| Knows your case | If the right person is available | No | Anchored to the case and the project |
| Applies the fix | You edit, recompile, re-run | No, you still do it | Proposes, applies, recompiles and re-runs |
| Who can do it | Someone who knows MXML | Anyone, with effort | Anyone on the team |
| Under your control | Yes | Nothing is recorded | Every change a proposal; every change in the Change Timeline, with undo |
03 WHAT IT MEANS FOR YOU
Faster fixes, still under your control.
It sees what you would have to open
The run's status, logs, failure message, failing line, screenshot and video, together, so you are not assembling the picture yourself.
A result, not a suggestion
It proposes a fix, applies it, recompiles and re-runs, so you see whether the case now passes rather than trying the idea yourself.
Under your control
In Ask mode each change is a proposal you accept or reject, and the Change Timeline lists every change, yours and the assistant's, with old and new values and one-click undo.
Your model, your keys
Claude, OpenAI or Gemini, on keys you hold, set per user or per project, so one project can trial it without committing the estate.
04 GETTING STARTED
Up and running in four steps.
- STEP 01
Choose your provider
Choose your provider in Project Settings and add your key. A per-project privacy setting governs whether the Assistant is available at all, and which models it may use.
- STEP 02
Start in Ask mode
Start in Ask mode , so every change is a proposal until your team knows what to expect.
- STEP 03
Try it on a failed run first
Open a broken case and press Fix with AI — the case is already not working, so it is the lowest-risk place to begin.
- STEP 04
Keep the Change Timeline open
Keep the Change Timeline open for the first week. It is the quickest way to build an accurate picture of what the assistant actually does.