There are two ways to adopt AI into how you work. The first is easy to grasp: fit AI into your existing process. You keep the steps you already have and lean on AI to move through them faster. That might be using deep research to kick off a task or Codex to tackle a specific feature.

This week's links are focused on the second option: changing the standard process itself.

Look at building a digital product. Traditionally that's some version of discovery → design → development. We front-loaded discovery and design because we had to be sure we'd identified the right problem and solution before development started. Otherwise we'd burn expensive dev time building the wrong thing. Because code was hard and time-consuming to produce.

As the cost of writing code drops, that sequence is up for grabs.

Development is still the last stop on the implementation train. But there's room to pull functional prototypes in much earlier. We can bring a demo app to clients instead of a slide describing it. Or build a rough version of a feature in the meeting where we're dreaming it up.

People make better decisions when they have more information. A prototype doesn't have to be perfect to help experts decide if an idea is worth exploring further. It's less about speed and more about being able to try more things. Ideas that used to hit the cutting room floor — too risky to build or too expensive to validate — are now cheap enough to test. That shifts the economics of our creative process, and experiments that never fit the budget or timeline are back on the table.

🤓 Read
Vibe coding and agentic engineering are getting closer than I’d like
Simon Willison
The article that set me off into the rabbit hole that ended up as the topic for this week. Simon Willison details his thoughts on how vibe coding and agentic engineering are starting to blend and what it means when you can suddenly produce 2000 lines of code in the time it used to take to output 200.

🎧 Deeper Dive on the High Leverage podcast

📺 Watch
Jenny Wen The design process is dead. Why Designers can no longer trust it.
Hatch Conference
Anthropic design lead Jenny Wen gave a presentation at Hatch Conference where she talks about how AI tools should impact the design part of the process.

📖 Read
The unreasonable effectiveness of HTML
Anthropic
I've been all in on markdown for everything when working with agents, but this article made me rethink that. An HTML artifact lets you create plans that can feature better diagrams and interactivity. The agent may use more tokens, but it'll give you an output that's going to be easier to understand and pass around to different stakeholders. I immediately used this to throw an interactive wireframe in a planning document.

📺 Watch
Inside YC's AI Playbook
Y Combinator
This was a very interesting dive into how Y Combinator has used AI tools in their production systems to become an AI Native company. They're basically capturing everything that happens in a database and have AI tools that everyone at YC can interact with. It's a very interesting concept that feels like the business level-up of Karpathy's knowledge bases.

📖 Read
Introducing dynamic workflows
Anthropic
Another one from Anthropic. This one on their new dynamic workflow feature. It basically lets Claude orchestrate its own agents to tackle complex problems in a systematic way. I expect we'll get similar features in Codex soon as that seems to be the release cycle. Just make sure you're ok burning some tokens.

Thanks for reading,
Jason