Menu

How I work with AI

I recently read Rich Holmes’s post Personal Agent Systems Explained - How 5 Leaders Run their work through AI Agents (well, the part before the paywall) and found it really interesting. I decided to ask Claude to make a diagram of my own system as well, and then I wrote some content around it, and now I’m sharing it with you in case it’s helpful (or you want to yell at me about how I’m doing it wrong).

How I Work is a mini site that goes through the whole thing, but here’s the overview diagram:

Diagram of my AI agent system

The main elements:

  • Me: nothing is fully autonomous. I don’t think that should ever change.
  • Context: what the agent knows about me and my work, loaded only when a task needs it.
  • Workflows: saved commands for repeated work.
  • Live data: agents check the real systems instead of guessing.
  • Review: a second model checks anything before it even gets to me for editing.
  • Lessons: corrections go back into the context, so agents don’t make the same mistake twice.

I’m sure lots of people have more sophisticated workflows and one of my reasons for posting this is that I want to hear about them!

The one thing that I don’t see a lot of out there, and that I want to point to specifically here, is how important it is for the system to be able to learn. The most important command I have is /session-end (generic version available here), which writes a handoff note and updates context and skill files. That way the system keeps learning and the agent never (well, rarely…) makes the same mistake twice.

Release: TL;DL v2.6.0 — Claude Opus 5.5 summaries and M4A support

Project
TL;DL
Summary
Your favorite podcasts, summarized.
URL
tldl-pod.com

Episode summaries are now written by Claude Opus 5.5, episode pages link to the episode itself, and long M4A episodes now transcribe. This release also cuts down on failed episodes from Substack-hosted shows.

Continue reading →

Release: listentomore v1.6.0 — Claude writes the summaries, pages load faster

Project
listentomore
Summary
Music discovery web app and Discord bot with AI-powered summaries.
URL
listentomore.com

This release moves the site's AI writing to Claude and makes pages that aren't cached load much faster. It also fixes summaries that were getting cut off or failing to load.

Continue reading →

An update from Amazon engineering leadership on the role of the developer in the age of AI

There’s not an awful lot new or surprising in this memo from Amazon leadership about the role of AI in software engineering, but it’s still interesting to see companies starting to codify the things that remain incredibly important for engineers (and product people!) to focus on. From real and continued first hand experience:

As a technical or organizational leader, part of your job is making sure that your team or organization is using AI to the best effect to improve delivery for customers and your business. That requires deeply understanding these technologies yourself, working with them hands-on rather than relying on what you hear, and understanding how others are using them. Sufficient understanding means you have used these tools on your production work, can describe a task where they helped and a task where they did not, and can explain that judgement to your team.

To the importance of staying close to customer needs:

We have also found that [Software Development Engineers] and [Software Development Managers] who are closer to their products and businesses are able to take the most advantage of AI and agentic software development. As Amazon has grown, we’ve seen some of our ICs and SDMs become less connected to their customers and business. We want to reverse this trend. We recommend that ICs of all levels find ways to become deeply connected to your customers, and your team or org’s product direction. As always, the scope and depth of this ownership will increase as you become more senior.

How To Write With An LLM

Don’t let an LLM pick your words. Be careful not to let it trick you into thinking your first draft is better than it is. Then outsource all the most tedious work to the model. Your voice stays intact, but your work is faster, better, and less painful.

—Thomas & Erin Ptacek, How To Write With An LLM

Slack Content Design in the Age of AI

I’m a words person from a UX background so I am deeply invested in how products communicate with users. This is a great article about how we should use AI to help with our content and not make it worse…

As AI takes on more of the work of generating language, the question is no longer whether machines can sound human enough. It’s whether those conversations earn trust, accurately set expectations, and reflect our human values. That gap is exactly where content design lives. With our deep expertise in applied linguistics, humanities, language systems, and behavioral economics, we are equipped to be the connective tissue between technical capability and the human experience.

I think the reduction of strong content people in organizations because “AI can just do the writing” is something we’ll regret even sooner than reducing software developers. Words, (I’m so sorry for the cliché), matter.

Why AI hasn’t replaced software engineers, and won’t

The research in this post is interesting, especially as it relates to where the software development bottlenecks are in the age of AI:

If writing code isn’t the bottleneck, what is? The task-breakdown surveys point at things like meetings or debugging. This just leads to more questions: what are developers doing in those meetings and why can’t it be done by AI? Won’t debugging get automated as capabilities improve? To understand the real bottlenecks, we have to get qualitative, and dig into software engineers’ own understanding of what it is they do that resists automation. When we did this analysis, it revealed three things as the real bottlenecks (1) deciding and specifying what to build, (2) verifying and being accountable for what is delivered, and (3) the deep human understanding — of the codebase, the business, and the environment — required to carry out both of these.

Beyond “good men”

This essay by Garrett Bucks about what’s going on at Cornell really got to me. I highlighted too many quotes, but here’s one that particularly stood out:

I’m writing this in the middle of the night. A few minutes ago, my nine-year-old daughter woke up and called out for me. She’s sick, and needed tissues and medicine. Of course I leapt to my feet. I’d do anything for her and her brother, such that they get to grow up enveloped in love and care forever and ever. We all would, us parents. At least that’s the story we tell ourselves. In practice, we accept a world where precious little is expected of men, and women are asked to fend for themselves. Boys will be boys, and girls will be given rape whistles and a laundry list of rules to follow.

How PM Work Is Changing in the AI Era

This article is a very good representation of how I see the PM landscape right now as well:

After a brief period where PMs were expected to become engineers themselves, the engineering team is moving back to owning most implementation mechanics. PM increasingly owns what should happen, why, the principles governing a product’s behavior, and how success should be evaluated. There are debates around communicating that vision between PRD’s vs. prototypes, but that debate hides the fact that the eval is becoming the core product artifact. The initial strategy doc or PRD is re-emerging as an artifact for agents as well as humans to understand what we intended to build. The eval defines whether we did or not.

Release: discogs-mcp v3.7.0 — Current MCP protocol, safer destructive tools

Project
discogs-mcp
Summary
Discogs MCP server.
URL
github.com/rianvdm/discogs-mcp

The server now speaks the current MCP protocol revision (2026-07-28), tells your client which tools can delete things, and asks self-hosters for one fewer secret. Older MCP clients keep working through the compatibility lane.

Continue reading →