Observations
1. A prototype-as-product model is replacing the build-then-dump cycle. At Anthropic, Steve says teams create many prototypes rapidly and just ship the best one. Claude Cowork reportedly went from prototype to launch in just 10 days. Meanwhile, “slot machine programming” – building 20 implementations and picking the winner – is becoming normal practice for teams.
2. The IDE could be evolving into a conversation and monitoring interface, not a code editor. Steve sees tools like Claude Cowork as the return of the IDE, focused on managing agent workflows above coding by hand. He predicts these new IDEs will focus on conversations with AI agents and monitoring them.
3. Reading ability is becoming a blocker for wider AI adoption. Some struggle with walls of text that current AI tools produce, and Steve predicts that in the very near future, most people will program by talking to a visual avatar, not reading terminal output because he observes that five paragraphs is already a lot to read for many devs.
4. AI coding has a spectrum, and most engineers trend near the bottom. Steve describes eight levels, from “no AI” to “multi-agent orchestration,” with most engineers currently at levels 1–2: asking an IDE for suggestions and carefully reviewing output. He suspects such engineers will be left behind.
5. Monolithic codebases are a big blocker to AI adoption in enterprises. AI agents have a ceiling of between roughly half a million to a few million lines of code which they can work with, effectively. If your codebase is a monolith that won’t fit in a context window, AI agents won’t work well with it.
6. What software engineers need to know keeps changing. In the 1990s, any decent software engineer knew Assembly, and today almost no decent developer knows it because Assembly has long been superseded by technical progress. What engineers “need” to know these days is different from the ‘90s and that process continues with AI, changing the parts of the craft that are essential for devs. We grumble about this but that won’t change anything by itself.
7. SaaS companies that don’t offer platforms and APIs will be out-competed. Steve uses Zendesk as an example: if your product doesn’t expose APIs, then AI-native companies will just build bespoke replacements. “If Zendesk doesn’t make themselves a platform, then they’ll put themselves out of existence.”
8. There’s a “Dracula Effect” where AI-augmented work drains engineers faster than traditional work because AI automates the easy tasks, meaning that engineers are stuck doing high-intensity thinking all day. Steve says you may only get three daily productive hours at max speed, but during that time, you could produce 100x more output than before.
9. Even if AI progress stalls, it’s worthwhile getting proficient at working with parallel agents. Steve argues that since there’s a model as capable as Opus 4.5 is, we don’t need smarter models but better orchestration layers. The worst outcome for someone who invests in learning AI tools is that they gain a skill set that stays useful, whether the models improve or not!