2026-10 - W1

The managers path in the age of AI

Deedy on X: “my favorite Paul Graham essay “how to do great work” in under 200 seconds of video(thanks opus)“

The Four Horsemen of Agentic Coding - by Alex Martsinovich

1. Slop: LLM-generated code has a strong smell that makes codebases repulsive to humans.
2. Alienation: Engineers become alienated from code and, as a result, care less.
3. Deskilling: AI erodes existing skills, impedes learning, and doesn’t offer any meaningful skill progression.
4. Team fallout: Agents weaken the social fabric of engineering teams by reducing interpersonal communication, increasing self-reliance, and hindering skill expression.

divenire on X: why i changed my mind about the death of software engineering

S always said experience in a well run corporate environment is invaluable, it will teach you things that you simply don’t learn gallivanting around early stage companies. the MO of early stage is spinning up shit as fast as possible and flinging it at the wall and seeing what sticks, but to build an empire, or even just be talent that’s hard to replace, you have to be knee deep in the mud and the grit and know how to navigate the tradespaces well.

LukeW | Design Systems for AI Agents

Sol and Aria’s websites follow the same approach (links go to their design system pages), each with design tokens, agent instructions, and a design system page that stays current because it’s built from the same code as the site.

Design teams have spent years trying to get people to actually read their design system documentation. Agents read it every single time.

You’re still the bottleneck | Ready, Set, Cloud!

What I am advocating for is not spreading yourself too thin. People have been saying for years that attention is one of the scarcest and most important resources we have.

In 2026, everything wants your attention. Humans have gotten really good at creating things that steal your attention away from something else. With coding agents, it takes almost 0 effort to bring in another work stream so you can “force multiply yourself.” Just open another terminal, type in claude and you’re (basically) done.

Protect it. Treat your attention like the scarce resource it is. Keep the decision-heavy workstreams to a minimum, just 1 or 2. Then run as many low-judgment streams as you want next to them. But keep your attention on the tasks that need it. Don’t split your attention across decisions, because when everything is a priority, nothing is.

Next time somebody asks you how many agents you have running, tell them that’s the wrong thing to be counting. The better question is how many decisions you’re actually making.

A programming language is a medium for people

There are two ways of thinking about LLM use in software development that I find useful to approach this discussion with clarity. The first has to do with the outcome of LLM-assistance and the second has to do with the process of using LLMs itself. … This is why LLMs are not compiler technology—it pains me that this need be said so often. Compilers operate on formal language where every aspect of program behavior is a known consequence of programming language semantics. Natural language however hides unspecified program behavior and LLMs don’t even reliably honor reasonably precise instructions. … A programming language is a medium for human beings. First, it conveys the intent and deliberation of a person over what they want their program to achieve for other people. Second, it is an empowering space for reasoning about and understanding program behavior—as opposed to machine code.

Kent Beck: Software Engineering in the Age of AI | Prodacity 2026 - YouTube

Nice talk from legendary Kent Beck!

Vibecoding isn’t as fun as writing code by hand — The Autodidacts

The word “dopamine” made me think: vibecoding is a bit like a series of first dates or one night stands. You don’t develop a relationship with the project: you don’t get the same kind of warm fuzzies, oxytocin, attachment. The problem with a dopamine-only life is that thrills become dull with repetition.

State of Devs 2026

AI Changed How Spotify Builds. What We Learned (and Fixed) About Quality at Higher Velocity | Spotify Engineering

Anti-Patterns in Software Blogging · Refactoring English

Awesome advice!

Jev-Driven SRE Diagnosis: What Worked and What Failed | SREGym Blog

Make Honesty Less Risky Than Silence for Your Engineering Team - ShiftMag

I like a simple communication profile for these conversations: POA – Polite, Open, Assertive.

</> htmx ~ Yes, and…

I don’t think programming is changing as dramatically as some people claim and I think the fundamentals of programming, particularly writing good code and controlling complexity, will be perennially important.

Some good sense here.

GitHub - zerobrewhq/zerobrew: An up to 6.6x faster Homebrew alternative · GitHub

Interesting. I should test it out someday.

The people holding up the internet | Data Drop

Agentic coding is a financial trap | Frugäast Blog

The current economics of artificial intelligence are fundamentally unstable. AI vendors have spent the last few years burning venture capital to subsidize model usage. As that subsidy ends, prices will inevitably rise. Corporate management often refuses to approve a $70/year license for a standard SQL productivity tool, yet happily approves spending massive amounts on enterprise-wide AI agent tooling. It is a misalignment of priorities driven by hype rather than unit economics. Until agentic harnesses stop relying on brute-force context sweeps and start employing actual semantic precision, operating open-ended agentic loops on a pay-per-token basis is a reckless allocation of your engineering budget. The ultimate luxury in software engineering isn’t infinite automation—it’s absolute, predictable control over your tools.

Fleeting — Sorry, I’m in a meeting

Funny simulated meetings.

Nix wrote half of my debugger | Farid Zakaria’s Blog

Interesting way of using nix.

Discovering Kindness: A Reflection on ‘Theo of Golden’

Recognition cost almost nothing A portrait of yourself is a strange gift. It tells you that someone noticed you and that you belong to this place. Most workplaces are bad at saying that. We celebrate launches and promotions because they’re visible. We say much less to the person quietly keeping everything running. … Mercy is attention under uncertainty. When you’re judging something someone did, you’re rarely seeing all of them, and this passage tells you what to do with the part you can’t see.

GenOffice — Free, Open-Source AI Office Suite

Alternative to office suites, with AI support.

Crafting Apps: open-source creative tools — ArtCraft

Alternative to Adobe’s tools. Interesting.

Five months treating bugs like patients and coding agents like a medical team

Using a medical team workflow, but on software world. Interesting, but not sure if that’s really better, as if I remember correctly, the medical team workflow is more like they don’t have any other choice to adopt this extreme workflow.

• We deployed a coding agent pipeline structured around a teaching hospital metaphor, using distinct roles like Triage Nurses, Fellows, Review Attendings, and Discharge Nurses, to enforce rigorous reviews, mandatory planning, and safety guardrails before merging code.
• Over a 5 month period, this agent pipeline successfully processed over a million lines of code with only 7 total reverts, proving that structured agentic workflows can safely build and maintain complex database migrations.
• By prioritizing meticulous record-keeping, strict test coverage, and planning reviews before writing code, the system minimized costly regressions and enabled features like the Migration Assistant, successfully balancing autonomous output with enterprise-grade software reliability.