From Autocomplete to Autopilot: How AI Is Changing the Developer Experience
From Autocomplete to Autopilot: How AI Is Changing the Developer Experience by Tugdual Grall
Room 8 · Deep Dive · 09:30-12:30
Summary
Section titled “Summary”This talk explores the rise of agentic workflows and context engineering in AI-assisted development. Using GitHub Copilot’s latest tools, it shows how teams delegate end-to-end tasks, from architecture scaffolding to PRs, and how developers, SREs, DevOps, and PMs shift from writing code to expressing intent.
Key points ⭐
Section titled “Key points ⭐”- 38:00 RTK, Many tool to save token
- 42:00 Hooks + Agent Debug logs (in VS code)
- 56:51 Tips use CLI in VS code
- 1:36:00 copilot impact explained
- 2:09:00 Rowork [?] setup
- 2:13:00 Project > setting -> Security -> codeQL
Timeline
Section titled “Timeline”- copilot CLI 10 people
- -> 500 PR/week
- quality + security needed (in CI ?)
- We want new feature -> so more PR
- But in a Bank App do you want new Feature every Day
- -> No you want security / stability
- Blog article Migration GHCP to Runtime Rust
- copilot chat Feb 23
- Agent Mode Feb 25
- Harness become better on Delta
- we have less need on skills Md file
- -> No need to explain How to write java (to be challenge)

- MR / PR code Review will be Most of our job
- CLI -> focus on Request
- -> Mult task
- copilot App -> focus on several task local ->

- cost Token cost on entreprise
- explain exemple $ 500 / users
- $ 2000 / users
- Demo start
- His credit since Begenin [?]
- +23.000 AI credit
- Shared experience on New Release of your AI tools
- go to the Release Note of the tool you are using
- RTK, Many tool to save token
- he only look at AI credit if too expensive use another Model
- do No early optimize
- personal plan => token
- entreprise plan => cost
- Hooks
- + Agent Debug logs (in VS code)

- Github.com
- -> configure Feature & client
- explain what Harness is Doing
- Tips use CLI in VS code
- Sandbox -> in copilot
- Autopilot => for code
- Manual => DB, PR Validation
- Auto pilot => Max AI credit
- explain before work tree for multi task
- Now he explaine Plan mode
- For Big Repo -> Say to look Big Module
- use the skill grill me
- skill.sh
- grill with doc
- Reduce the spec kit to use plan
- Feature Flag important
- copilot impact explained
- I should manage this
- But maybe only valuable if using GH infra
- code completion not interesting

- ruber duck Feature GHCP
- to use another model for Review
- Demo use Astro 8 000 AI credit
- to Rewrite the Full pet clinic
- Side chat
- New chat with the same Branch
- luna max [?]
- Rowork [?] setup
- He has a prompt take pet clinic
- Rowork [?] Back end , Front end
- same prompt + same cortex [?]
- with ≠ Model
- have a bench mark to know New model ⭐
- his point if you Diolant [?] change Agents .MD, skill, context since 6 Month you miss something
- thus change so Fast
- Project > setting -> Security
- -> codeQL
- Code QL can be used for SFPD Not just security
- /Review
- -> is it something we can use in local ?
- /chronicle
- -> Tips
- -> improve (Defined Rule)
- still need validation and not fake improvement.
- Gain insights across your agent sessions with /chronicle - GitHub Changelog https://github.blog/changelog/2026-06-02-gain-insights-across-your-agent-sessions-with-chronicle/
- awsome copilot
- on size fit everything doesn’t work
- skill was nice 6 month ago
- less today
- may be Back in 6 month
- create skill, agent But Basic
- Touri [?]
- -> use AI to learn
- /spor [?] => Challenge me
- Run 2 agent for evaluate each Tools
- the Compare the Result
- if the prompt is not Big enough (Not Describe)
- it should Be a plan
- ask GHCP Review work on ≠ session
- he can do that with the “chronicle Dalan [?] Base”
- Add jira canvas [?] to add user interka [?]
Extra notes
Section titled “Extra notes”- When chat is the wrong UI - The GitHub Blog https://share.google/AOIopGe27Bds7UEB0