AI state of the union: Reflections Part 1
10 Apr 2026
Reflections whilst watching: An AI state of the union: We’ve passed the inflection point & dark factories are coming with Simon Willison.
00:00 to 20:00.
00:00 to 20:00.
- November 2025 - the inflection point where OpenAI's GPT 5.1 and Anthropic's Claude Opus 4.5 crossed a threshold. This app building stuff actually works now. By Jan and Feb 2026 folks realise this actually works now.
- Can churn out 10k lines of code in a day, how do we get to "most of it works" to "all of it works"?
- Vibe Coding. You don't look at the code. Hands off. Was good for prototyping. Yet now it's for non-programmers. We have democratised the art of getting a computer to do something for you. Vibe coding for single use users is fine.
- The moment others will be using a vibe coded app is where it gets dangerous. "This is not a responsible way of using these tools."
- How do we know what is responsible and what isn't? Many ways to cause damage if people don't know what they are doing.
- Agentic Engineering is a better term for vibe coding for software professionals who are doing responsible vibe coding.
- How do we build professional software using coding agents? Let's not just build good but better than we were building before. Let's go for higher quality software.
- Dark Factory pattern or Software Factories. Currently a professional tells AI agent what to build, carefully review code and make sure it's doing the right thing. However with a Dark Factory what does it look like if you're not reviewing the code but you're also not vibe coding? You're applying professional practices and quality expectations to code that you're not directly reviewing. You can "turn the lights off" – like you can do in an actual factory when all the machines are operated.
- Todo: See what the company StrongDM are up to.
- Some companies are implementing a "Don't write code" policy.
- The next rule would be "Nobody reads the code".
- How do you produce software that works and is good if you're not reading the code?
- StrongDM were simulating testers. They setup agent QAs to act like users. Everything was simulated via a Slack.
- Cost tens of thousands a day for tokens. They had 24 hours of a swarm of testers.
- Because of restrictions on SaaS tools, they built a simulation of Slack and Jira and other software. Took APIs and told coding agents to build them.
- Even if your virtual QA team says its good it doesn't mean it's secure.
- Agents are getting good at security penetration testing.
Simon Tomes
Community Lead at MoTaverse
he/him
Hello, I'm Simon. Since 2003 I've had various roles in testing, tech leadership and coaching. I believe in the power of collaboration, creativity and community. 🎓 MoT-STEC qualified.
Sign in
to comment
Manage your entire QA lifecycle in one place. Sync Jira, automate scripts, and use AI to accelerate your testing.
Explore MoT
Thu, 1 Oct
Previously known as TestBash, MoTaCon is the new name for our annual conference. It's where quality people gather.
Boost your career in software testing with the MoT Software Testing Essentials Certificate. Learn essential skills, from basic testing techniques to advanced risk analysis, crafted by industry experts.
Debrief the week in Quality via a community radio show hosted by Simon Tomes and members of the community
