An AI state of the union: We’ve passed the inflection point & dark factories are coming
Simon Willison, co-creator of Django and veteran engineer with 25+ years of experience, argues that AI has crossed an inflection point in late 2024 where coding agents now reliably produce working code—shifting the bottleneck away from code generation to ideation, testing, and maintaining quality. The episode explores how this changes what software engineering means, why mid-career engineers are most at risk, and how the emerging "dark factory" pattern could enable production software built and tested entirely by AI agents without human code review. For builders, the key insight is that code is now cheap, which means the winners will be those who use this velocity advantage to ship higher-quality, more ambitious products rather than simply faster output.
Key takeaways
- • Coding agents have crossed a threshold where they "almost all of the time" do what you tell them to do—meaning you can now prototype three different solutions to a design problem and test which one users prefer, shifting product development from "build one thing well" to "explore multiple directions cheaply."
- • Test-driven development is crucial when working with coding agents because it forces agents to run and verify code; teams removing tests for speed are making a mistake, as tests actually accelerate development by preventing regressions when adding features.
- • Building a personal knowledge base of working code snippets, prototypes, and research projects (stored in GitHub or similar) lets you instruct agents to combine prior solutions—Simon maintains 193+ small HTML/JavaScript tools and 75+ research projects that agents can reference to solve new problems faster.
- • Mid-career engineers are at highest risk from AI disruption (not juniors or seniors): experienced engineers can amplify their skills with agents, interns onboard faster, but 5-15 year engineers lack both the deep expertise to leverage agents effectively and the fresh perspective of AI-native developers.
- • The dark factory pattern—where software is built and tested by agents without humans reading code—is already being tested in production; StrongDM uses swarms of simulated end-users running 24/7 testing against simulated APIs, spending $10,000/day on tokens to validate security-critical software without manual review.
- • Mid-career professionals should lean into AI as a learning amplifier: use it to prototype in unfamiliar domains you'd normally skip (like AppleScript), maintain ambitious projects, and build agency by deciding *what* to build rather than *how* to build it.
Recommendations (23)
"Anthropic came up with Claude Code back in sort of February of 2025 and it took off like crazy and a bunch of people started signing up for $200 a month accounts"
Simon Willison · ▶ 3:45
"he's been sharing everything he's learning in real time through his incredible blog, simonwillson.net."
Lenny's Podcast · ▶ 2:20
"I never used Apple Script because Apple Script is a whole programming language you have to learn and I've been using Apple Script for like two and a half years now because ChatGPT knows Apple Script"
Simon Willison · ▶ 32:09
"I use that one more than the one on my own computer. Partly because that's the one you can access through your phone."
Simon Willison · ▶ 48:52
"If you've got the Anthropic Claude app installed on iPhone, there's a code tab and you can go in there"
Simon Willison · ▶ 48:59
"I think it's on par with Claude Opus 4.6 and possibly even better."
Simon Willison · ▶ 51:01
"I've got 10,000 Apple notes as well that I just constantly add new things to"
Simon Willison · ▶ 1:05:09
"So I told Claude Opus, I said, 'Here is the code for the OCR, the PDF thing I did. Here's the code for the OCR thing. Build a new thing that can open a PDF file and OCR every page.' And it did it."
Simon Willison · ▶ 1:07:12
"This is the most important thing when you're working with coding agents is they have to test the code. The best way to do that is to use a programming technique that we've been using for decades ca..."
Simon Willison · ▶ 1:08:57
"If you use the term red/green TDD, that's programming jargon which I didn't used to use, but it is jargon for run the test and watch them fail. The agents know what that means."
Simon Willison · ▶ 1:12:13
Mentioned (20)
More from these creators
The tactical playbook for getting 20-40% more comp (without sounding greedy) | Jacob Warwick
How I built a 1M+ subscriber newsletter and top 10 tech podcast | Lenny Rachitsky
The most successful AI company you’ve never heard of | Qasar Younis
The design process is dead. Here’s what’s replacing it. | Jenny Wen (head of design at Claude)
AI is critical for humanity’s survival: Cisco President on the AI revolution | Jeetu Patel