When code is cheap, judgement becomes the job (opens on the source site)
Software engineering has changed in the last 2 years. Here's how that looks without the fluff. Real company, big stakes, finite token budget
Read at the sourceTHE COMPANY INDEX TRACKED BLOG
Swizec Teller
Ideas, decisions, and lessons from the team.
swizec.com (opens on the source site)21Posts tracked
3 days agoLatest publication
1.8Posts / month over the last 12 months
Latest writing
20 of 21 postsWe now auto-approve and merge 15% of PRs (opens on the source site)
've been on a quest lately to reclaim my sanity at work. The latest advancement is a bot that auto-approves and merges low risk PRs with no oversight 😈
Read at the sourceWhat B2B SaaS taught me about having kids (opens on the source site)
I've been a dad for 3 months now so obviously I have it all figured out and there's nothing new to learn. My theories are rock solid and definitely apply to everyone.
Read at the sourceFocus and followthrough are the moat (opens on the source site)
The points are fake and the score is made up but we always finish 70% of our sprint and get around 30 points done per person. Hey waitaminute that almost sounds useful 🤨 I've said before that working in a high growth startup is like spinning plates – you do a bit here a bit there and kick the can to survive another day. There's always more work than people and spaceship-first development makes this worse because starting is easy and finishing is hard. On my team we add ~50 new ideas to triage every sprint, work on 5 or 6 ongoing major projects, keep our production systems running (the SRE…
Read at the sourceStop burning tokens on code review (opens on the source site)
I've been experimenting with different approaches to build a system where humans and agents can ship fast safely. Think I've found something that works – custom linters.
Read at the sourceHow to be useful as a software architect (opens on the source site)
Software architects have a weird role: Think about the code _tomorrow_. Today's code is what it is, how do we make it good tomorrow? Without slowing down, losing business, or falling off a cliff.
Read at the sourceThe one thing every experienced engineer gets wrong (opens on the source site)
In the last few months every engineer I hired stumbled in exactly the same way: Project management. They learned quickly, but that first big project has become a rite of passage.
Read at the sourceYou get what you hire (opens on the source site)
Hiring engineers is like deciding what kind of system you'll have. This seems obvious but surprised me to see first-hand.
Read at the sourceCode is the easy part, or how we refactored half the business to fix a janky script (opens on the source site)
This is a classic startup war story: Someone had built a leaky rowboat. It worked! Then we started flying the rowboat as a Cessna. What we needed was an aircraft carrier.
Read at the sourceGood software gets out of the way (opens on the source site)
Explore the art of invisible software and user-centric workflow design with Swizec and Kent C Dodds. Uncover what separates a good engineer from a great one in this insightful dialogue.
Read at the sourceUser outcomes, workflow design, and biotech software (opens on the source site)
Kent C. Dodds talks with Swizec Teller about product engineering for software that serves real businesses and non-developer users.
Read at the sourceDear [Software Engineering] (opens on the source site)
Clara Shih talks with Chip Huyen and Swizec Teller about what entry-level software engineering looks like in the AI age.
Read at the sourceRelentlessly focus on the outcomes (opens on the source site)
AI makes it easy to focus inwards and feel impressed with all the churn and motion and output you're producing. But your users don't care.
Read at the sourceEngineering levels aren't about skill anymore (opens on the source site)
Traditional engineering levels were about skill progression. I think this has changed. Everyone now is a mini staff engineer.
Read at the sourceJust fix their PR (opens on the source site)
You know what's better than a code review comment? A fix. Lately we've been dealing with an absolute onslaught of pull requests. Feels like half my days go to reviewing code, half to management, and half to writing code. Yes that's too many halfs. See I wasn't even exaggerating (thought I was). 53% of my contributions are code review 🫠 Part of this is because AI now writes 97% of my code. I review that and like to leave the bot comments for things to fix. Feels natural and I don't have to learn a new workflow. Oh a human wrote this code? Here's a comment, please fix. Oh a bot wrote this…
Read at the source"But if AI does it all, what's *my* job?" (opens on the source site)
With all the AI stuff one day you feel useless, another day you feel top of the world with an army of agents building big things that used to take weeks. And then your brain melts trying to keep track of it all. You own too many domains, are always on-call, and if anyone wants anything you are the person to ask. Because who else? You built it, you own it, you run it, you never told anyone else how it works. You barely looked at it yourself. Welcome to the "mental load" thing feminists have been talking about for ~~years~~ decades. Managing the work is harder than being told what to do and…
Read at the sourceWhat you need to make an agent (opens on the source site)
I've been digging into agents for internal tooling and processes. Looking for how to scale without drowning in manual work. Here's what I found
Read at the sourceQuick note on evals and putting AI in your resume (opens on the source site)
When candidates put AI on their resume, the key thing I try to find out is whether they used evals. How did you measure making improvements?
Read at the sourceAI now writes 97% of my code. Here's what I learned (opens on the source site)
This week I was shocked to learn that AI now writes 97% of my production code. The stats come from Cursor and they're not incentivized to lowball, but honestly looking at recent weeks ... yeah that's about right. This is not a toy project Now, I'm more of a manager these days than a heads down code grinding IC, but I own production features with big stakes and real money attached. We're an almost billion dollar company and some users pay 6-figures in a single order. This is not a toy indie hacking project with a few dozen users. To give you a sense of scale: my most recent big project was…
Read at the sourceWatch people work (opens on the source site)
We talk a lot about outcomes over outputs and how engineers are supposed to get us over the water, not blindly build a bridge because someone asked for a bridge. But how do you do that in practice?
Read at the source