Thorsten Ball's Agentic Coding Vision
Table of content

Thorsten Ball is a German software engineer who writes books, builds editors, and thinks publicly about programming. He wrote Writing An Interpreter In Go (2016) and Writing A Compiler In Go (2018), both self-published books that teach by building working systems from scratch. Before moving into AI tooling, he worked at Zed, the GPU-accelerated code editor built in Rust.
Ball now leads development of Amp at Sourcegraph. His Register Spill newsletter documents his evolving relationship with AI coding tools.
The Tension
From Ball’s Joy & Curiosity #66:
“Friends of mine — very experienced and very good programmers — are saying loudly and publicly and in private too that agents are useless. It’s slop, they say. It’s all averages. No brilliance, no creativity, and half of it doesn’t work.”
Yet his personal experience differs:
“I saw Amp knock out stuff that would’ve taken me days and, as a junior engineer, possibly weeks, if I had done it myself. I saw Amp build a tiny and brilliant renderer for box-drawing characters in my terminal emulator.”
Background
Ball’s career spans:
| Period | Work |
|---|---|
| 2014-2019 | ioki (mobility startup) |
| 2019-2024 | Sourcegraph (code intelligence) |
| 2024 | Zed (code editor) |
| 2024-present | Amp at Sourcegraph |
He joined Zed in January 2024, where he conducted interviews with the founders about their technical choices and product philosophy. He then returned to Sourcegraph to lead Amp development. His blog includes posts like Professional Programming: The First 10 Years and How can you not be romantic about programming?.
How I Use AI (March 2024)
Ball’s How I Use AI documented his daily workflow before leading Amp:
“Occasionally I come across comments that make me think ‘wait, there’s still developers out there who don’t use AI? Like, not at all?’”
His stance: use AI pragmatically without strong opinions about whether it’s “good” or “bad.”
Use it like Google, StackOverflow, or any other tool.
Say "thank you" just in case.
Amp and Sourcegraph
Amp Code is Sourcegraph’s agentic coding tool. Ball returned to Sourcegraph specifically to work on it.
From the Latent Space podcast (September 2025):
- Sourcegraph ships 15x/day with no code reviews
- Subagents and prompt optimizers “aren’t a promising direction”
- Focus on rapid iteration over complex architectures
Key product philosophy: fast deployment and tight feedback loops matter more than sophisticated agent orchestration.
Agentic Coding Changes Everything
Ball’s July 2025 Changelog podcast appearance titled “Agentic coding changes everything” summarizes his position:
“Being able to talk to an alien intelligence that edits your code changes everything.”
Topics covered:
- How coding agents actually work
- Recent AI tooling advancements
- Amp’s approach vs competitors
- The divide between believers and skeptics
Pragmatic Engineer Discussion
Gergely Orosz interviewed Ball for The Pragmatic Engineer (October 2025). Key topics:
- Programming language comparisons (Python, Go, Rust, TypeScript)
- How AI changes language choice considerations
- Ball’s evolved perspective on AI tooling
Quote: “I’ve changed my mind about AI tools.”
Writing as Thinking
Ball’s books on interpreters and compilers reflect his approach: understand systems by building them from scratch. He wrote about the process in Why I Wrote a Book About Interpreters and What I didn’t do to write a book. His AI journey follows the same pattern: build the tools, ship with them, form opinions from experience rather than theory.
Key Insights
| Observation | Implication |
|---|---|
| Experienced friends dismiss agents | Social proof fails for emerging tools |
| Personal results contradict skeptics | Individual experimentation matters |
| Ship 15x/day, no code reviews | Speed beats ceremony |
| “Alien intelligence” framing | Agents are collaborators, not autocomplete |
Links
- Personal Website
- Register Spill Newsletter
- Blog
- Amp Code
- How I Use AI (March 2024)
- Joy & Curiosity #66
- Changelog: Agentic Coding Changes Everything
- Tuple Podcast: Thorsten Ball at Zed
- Writing An Interpreter In Go
- Writing A Compiler In Go
- GitHub: @mrnugget
Next: Geoffrey Litt’s Malleable Software Vision
Get updates
New guides, workflows, and AI patterns. No spam.
Thank you! You're on the list.