Hi Alex,
thanks so much for giving this topic attention — I truly appreciate it!
Since my post last year, the space has evolved quite a bit, and through hands-on work (mostly from Python-based side projects), I’ve developed a clearer view of what a truly top-notch developer experience could look like in this domain.
More than half of the following I can do today in a OpenAI ChatGPT webinterface to their models.
Even if Studio just supported me in getting “context” out of it that would be highly beneficial. For example an export function of the project variables and arguments in the panel “Data Manager” that I could paste into a chat.
Refined RPA Copilot Developer Assistant Requirements, 2025-Q2 edition
- Multi-Threaded Chat per Project
- I’d love to have multiple Copilot threads within a single project.
- Each thread would keep its own dedicated context (e.g., coding, researching process patterns, git guidance).
- This would really help with mental and functional separation — much like having multiple tabs in Studio, but purpose-built for the Developer Copilot.
With web frontends to chat assistants I repeatedly open new chats, actively managing context.
-
LLM Switching per Thread
- I’d find it valuable to choose the most appropriate LLM for each thread (e.g., GPT-4 for code-heavy tasks, Claude for broader reasoning).
-
Context Enrichment via Inputs
- I’d like to enhance each individual chat context by:
- Dragging in PDFs
- Referencing local folders
- Supplying URLs to draw in external domain knowledge
With web frontends to chat assistants like ChatGPT custom GPTs I do that all the time, even if I just create a custom GPT in draft mode and use the preview chat
- Pseudocode & Code Generation
- The Developer Copilot should be able to generate UiPath-style pseudocode (e.g., “Use Excel Process Scope → Read Range → For Each Row”).
- It would be great if it could optionally produce
.xaml
fragments for manual insertion (e.g., via Clipboard AI-style copy/paste).
- More importantly, I’d strongly prefer that the Copilot prioritizes coded automation generation (see Studio - Creating hybrid automations).
From my perspective, `.xaml` as a target format just isn’t viable for LLMs:
- It’s too verbose, which bloats context
- my rule of thumb in Python/JavaScript/PHP coding chats: Max 500 lines of code, beyond that the LLMs output quality is deteriorated
- It only allows static reasoning
- The actual logic is buried in escaped attributes and dynamic bindings
In short, trying to generate .xaml
directly feels like a dead end — and with Coded Automations already available, I think the transition path is clear: DROP xaml.
- Auto-Documentation in Markdown
- I’d like to have Copilot generate a Markdown “wiki page” in a
Documentation/
folder when I ask it to summarize a topic or discussion
- A simple built-in Markdown viewer in Studio would complete the loop nicely
Generating contextual README and wiki pages is truly mindblowing! GitHub does it well with the complimentary wiki repositories (for a repository foo-bar one can have a foo-bar.wiki repo that is shown in the repo webgui).
- Software Quality Coaching
- The Developer Copilot could support best practices by:
- Suggesting relevant tests
- Recommending strong commit messages
- Promoting consistent folder structures and naming
- Reminding me when I overlook these steps
That is then a real assistant! (e.g. “You changed 3 files, and now are asking about a new topic, you should commit first”).
This is actually not a nice-to-have feature but a must-have: Generated code is expected to break the current implementation, and a rollback to previous version has to happen frequently. Modern AI-centric IDEs require rollback features like never before.
- Environment Awareness
- The Copilot should understand:
- Installed packages
- and whether updates would resolve known issues(!)
- Existing variables and arguments
- Project structure overall
- It could help ensure clean naming and suggest proper use of patterns (e.g., putting data into
TransactionItem.SpecificContent
)
The release notes of the avtivities are a top priority to get into context.
- Debugging Assistance
- I’d love help with:
- Understanding error messages
- Knowing where to place breakpoints or add logs
- Getting interactive suggestions for likely causes and fixes
Analog to “extract as image” I would like to “wrap in testcase”.
-
Change Impact Analysis
- If I make a change, Copilot could flag:
- Dependencies that may be affected
- Workflows that might break as a result
- That would be a huge help in reducing regressions
-
Pattern Recognition & Reuse Suggestions
- The Copilot could detect repetitive logic and recommend:
- Creating reusable libraries
- Refactoring into cleaner sequences or workflows
The current approach I see in “Autopilot for Everyone” in UiPath Assistant is not convincing to me for the reasons
- not project-files aware!?! The code assistant knows NOTHING about my project files, not even those I have open in the Studio!? That’s the first thing e.g. VS Code plugins for GitHub Copilot did, back in January 2023
- context / knowledge files are administrated in the Orchestrator Tenant where I will most likely have no permissions – and it is too far away from the task at hand.