AI Agents

Ars Technica primer on AI coding agents

If you've been curious about AI coding agents (tools from OpenAI, Anthropic, and Google that can work on software projects for extended periods, writing apps and fixing bugs), this Ars Technica piece offers a clear primer. At their core, coding agents are program wrappers that orchestrate long-running work done by large language models: a supervising layer interprets tasks from humans, decomposes them into steps, and manages calls to tools that can write files, run commands, and execute code. This structured control layer is often referred to as the agentic harness.

A key constraint is that every LLM has a limited context window—essentially short-term memory—that fills up as interactions grow. To work around this, agents use techniques such as context compression (summarizing prior activity to preserve key details while discarding redundancy) and external documentation or memory stores that allow the system to persist state and reorient. Multi-agent architectures can tackle complex problems in parallel, but they consume tokens far more aggressively than typical chatbot use—often by an order of magnitude or more.

For anyone managing technical teams or working alongside developers, understanding these mechanics helps set realistic expectations about what AI coding agents can and can't deliver—and why human planning and oversight still matter.

Ars Technica | How AI coding agents work—and what to remember if you use them
☕ Medium Read (7 minutes)

AI Agents

What AI coding agents can actually do now

Ethan Mollick's latest piece is an excellent follow-on to the Ars Technica primer above. His post demonstrates how recent developments have greatly expanded what AI coding agents can do:

"I opened Claude Code and gave it the command: “Develop a web-based or software-based startup idea that will make me $1000 a month, where you do all the work by generating the idea and implementing it. I shouldn’t have to do anything at all except run some program you give me once. It shouldn’t require any coding knowledge on my part, so make sure everything works well.” The AI asked me three multiple-choice questions and decided that I should be selling sets of 500 prompts for professional users for $39. Without any further input, it then worked independently… FOR AN HOUR AND FOURTEEN MINUTES, creating hundreds of code files and prompts. And then it gave me a single file to run that created and deployed a working website (filled with very sketchy fake marketing claims) that sold the promised 500 prompt set."

I highly recommend reading this post in full. As Mollick concludes:

"If you're programming-adjacent (an academic who works with data, a designer who wants to experiment with code, anyone who wants to try building a thing they are imagining), this is your moment to experiment. But there's a deeper point here: with the right harness, today's AIs are capable of real, sustained work that actually matters, and that, in turn, is starting to change how we approach tasks."

Ethan Mollick | Claude Code and what comes next
☕ Medium Read (10 minutes)

More light-hearted (but impressive): Casey Newton and Kevin Roose, hosts of the Hard Fork podcast, share the "over the holiday" software projects they created, as non-coders, with Claude Code—and why Claude Code has been having a moment. Segment starts at 27:00.

Product Development with AI

The shrinking middle of software development

Karri Saarinen (co-founder of Linear) argues that software development is hollowing out in the middle. For decades, the bulk of the work lived between "idea" and "shipped"—the actual coding, debugging, and implementation. With coding agents increasingly handling that middle layer, the valuable work shifts to both ends: defining what should be built (the intent, context, and constraints) and evaluating what comes out (reviewing, testing, and ensuring the output actually solves the problem).

In Saarinen's framing, design becomes less about producing artifacts and more about clarifying intent—understanding what matters and what trade-offs are acceptable. It's a useful mental model for thinking about where human judgment remains essential as AI takes over more execution.

Karri Saarinen | The disappearing middle of software work
⚡ Quick Read (2 minutes)

Related: Simon Willison on why coding agents make it no longer remotely excusable to skip out on quality engineering processes

AI Agents

Claude Code... for everyone else

From Anthropic:

"When we released Claude Code, we expected developers to use it for coding. They did—and then quickly began using it for almost everything else. This prompted us to build Cowork: a simpler way for anyone—not just developers—to work with Claude in the very same way. Cowork is available today as a research preview for Claude Max subscribers on our macOS app, and we will improve it rapidly from here. How is using Cowork different from a regular conversation? In Cowork, you give Claude access to a folder of your choosing on your computer. Claude can then read, edit, or create files in that folder."

This is a good recognition — and leverage — from Anthropic for the non-coding use of Claude Code. Cowork does not require the terminal; it's all within the Claude experience familiar to most Claude users. Definitely check out the demo videos!

Introducing Cowork: Claude Code for the rest of your work
⚡ Quick Read (2 minutes)

AI Agents

The case against app-by-app AI agents

In closing, there’s this take from Ethan Mollick:

"Microsoft’s (and many other traditional software vendors’) bet that people will want each app infused with its own focused AI is looking like a bad one in the face of Codex, Antigravity, and Claude Code. People like to delegate to an agent that works across apps to do tasks."

In this thread, he specifically calls out Google as an example:

"The scattered Gemini implementation (every app in Google’s ecosystem uses it differently) is standing in the way of Google actually creating a single interface for non-coders to do agentic work. They have the pieces, but they haven’t been put together yet."

Ethan Mollick | Thread on X

That’s it for this week.

Thanks for reading, and see you next Wednesday with more curated AI/UX news and insights. 👋

All the best, Heidi