Taimoor - 🦀👨‍💻

I devoured the recap @martinfowler posted from the deer valley summit. Loved it. But the notes suggest we may be replicating a perennial blind spot in software engineering: treating code like the outcome, and production like an afterthought. honeycomb.io/blog/productio…

I give it less than 6 months before Garry stops preaching LOC and starts preaching maintainable code bases. And with that one move he will go from junior engineer to a bit more senior. We watching his Eng journey live 🍿

In an attempt to fight "slop creep" (thanks @boristane for the term) I tried more conscious smaller scale changes today. The main learning from that: if the changes are sufficiently small, there is little to no difference between how Opus and Codex behave.

This is the crux of the Cybersecurity going forward - "Monitoring is the only stage of the SDLC that survives. And it doesn’t just survive, it becomes the foundation everything else rests on." We need to safely mutate code as quickly as we find vulns. boristane.com/blog/the-softw…

@GergelyOrosz Idk if it even depends on the size of the company. MCPs expose tools to AIs. Sentry has mcp which can connect to any agent and I can work with it to solve issues. Sonar's mcp helps agent find and correct quality issues. How else would you've done these??

MCPs are the opposite of dead. They are the life blood of how AI agents use services inside mid-sized and above companies. Case in point: Uber runs on MCPs internally, for good reason. Details: newsletter.pragmaticengineer.com/p/how-uber-use…

Last week a vendor tried to twist our arm. Today we switched to a vibe-coded replacement. Wild. If you're in enterprise sales, be careful not to overplay your hand. The power balance is shifting towards buyers.

Why do we so often treat refactoring as a "nice to have" rather than a core engineering discipline? We've all been in situations where we dread opening a specific file because the design is poor, obtuse, and overly complex. But the value of refactoring... 🧵 1/7

genie is out of the bottle everyone hitting the magic button button puts your brain in a state of laziness that seeps into all your processes and they get skipped we talk about this all the time and still our team is struggling with it

@GergelyOrosz We have a GitHub copilot subscription that we use with opencode and the copilot-cli sometimes. It's great because it gets us frontier models without lock in

Actually, I do hear more startups “taking away” GitHub Copilot from devs - and no one is complaining at those places. Because those devs don’t use Copilot, and are on tools like Claude Code, Codex, Cursor agents etc. So companies just cancel the unused Copilot licenses.

Saying not knowing how to code gives you an advantage in building software (thanks to AI) is like saying not knowing anything about filmmaking gives you an advantage in making films (thanks to having a smartphone + apps to edit stuff) Ignore this stuff and keep learning+building

@faisal_aftab We're doing our best at Blended Labs get-blended.com

We have a short window to rethink K-12 and higher education before AI disruption fully hits the system. Anthropic just introduced “observed exposure,” comparing AI’s theoretical capabilities with real usage data from Claude to measure its real impact on jobs. Link: the-decoder.com/anthropics-new… #professions #education #jobs #ArtificialIntelegence @AnthropicAI

@nbevans Why is it a mistake? Should've stayed with the tooling focus?

"At the all-hands, Cursor leadership warned that the months ahead would be turbulent ones. Projects might be scrapped, priorities shifted. The company’s new mandate was labeled “P0 #1”—priority zero: “Build the best coding model.”" Cursor just made a big mistake with this pivot

@MarcJBrooker But the lines are blurry tho for some cases: x.com/i/status/20297…

This space is moving so fast, and so it's natural that folks can't keep up.

The MCP vs CLI discourse is weird, because it's all missing the point. What MCP is good at (integrating with remote tools, services, and databases; dynamic discovery; etc), and what CLIs are good at (local, dynamic data but static tools, etc) are mostly independent.

@nbevans @MarcJBrooker Sonar's mcp runs locally in docker. I got rid of my custom scripts with it. GitHub has mcp. But I never set it up because i think the agent just uses the gh command very well. Idk how to reason about the pros and cons

@MarcJBrooker Yep, they're both good at different things CLIs are more powerful though but the drawback is you need to install shit locally which might be a deal breaker in enterprise environments

The conversation around coding agents has evolved significantly since December. Despite the early debates and attacks, the widespread endorsement all the way from Knuth to Torvalds grounded things. I’ve been just doing my best to be intellectually honest given the new generation of programmers deserve to understand what’s happening.

I keep seeing this pushback on AI coding: "I love the intellectual challenge of taking something that only exists in my head and making it real." I fee the same. That feeling never gets old. And I don't think AI takes it away from us. If anything, it gives us more ways to make things. I used to avoid anything frontend. Not because I didn't have ideas, but because the gap between what I could imagine and what I could build was too wide. AI closed that gap. Now I ship things I never would have attempted. The creative joy isn't gone. It's just no longer gated behind skills I don't have - UX design - or skills I don’t have the desire to learn - centring a div!

@johnzabroski @fbrasisil @headinthebox What's the alternative? Haskell? Or do you mean going for popular languages like Typescript or Python?

@fbrasisil @headinthebox But why Scala? As much as I love zio, the reality is the type overhead eats away at performance. Why create the type overhead if, as Erik asked, the agent does all the scaffolding for you? Feels a bjt like wanting to drive stick shift to the point of refusing to drive electric !

A nice plot twist: Scala's bad DevEx is greatly mitigated by AI agents. I don't bother with broken IDE refactoring, don't need to fight sbt anymore, don't need to debug stuff because of bad docs, can easily pinpoint compiler bugs and find workarounds, etc. Just ship 🚀🚀🚀

Remember good old Fred Brooks: “Adding manpower to a late software project makes it later.” We need to update it for the AI era: “Adding manpower inferior to AI makes a late project even later.” My own small state university has over 50 tech folks. With AI, how many do they need? They run our Oracle database where grades and metadata are stored. They house course material. They run our website. And they manage a fleet of laptops and basic services like an internal Cisco network (barely used now that all staff work from home). They used to run an email server, but this was outsourced to Microsoft some time ago. They used to install Microsoft Office on machines, but this is also outsourced. Most universities in Quebec have moved their servers to the cloud. Our IT people do occasional custom software development, but it is extremely slow (typically years). Not because they are incompetent, but because of the culture. For the most part, none of them do bleeding-edge computing. Not that they could not intellectually, but that’s not the job. By my estimation, three people with AI could do the work of these 50 people. “But Daniel, aren’t you an AI skeptic? Aren’t you on record as saying that there won’t be mass unemployment?” Yeah. And my scenario of three people replacing the 50 won’t happen. Not in the 2020s, not in the 2030s. Furthermore, the smartest folks among these 50 people can do more in the future. We are absolutely limited by computing: we have countless processes that are email-heavy while they could be largely automated. We fill out Excel spreadsheets that are then hand-processed by staff members. And do I need to tell you how much in higher education can be improved by AI? The future is already visible: all students will have private AI tutors, all-knowing, available 24/7. So there is plenty of work. But not everyone is “as good as AI.” We have spent the last 15 years growing the size of tech staff and, at the margin, we have added many people whose interest and innate abilities in computing are low. Nobody will get fired at my university. And they will keep growing their base for quite a long time. But new organizations will pop up and they will be structured differently. My prediction is that a lot of high-agency smart young people will pick up jobs where they run systems that used to require 20 people. And they will seek high compensation and high status. And mostly, they will get it. It will increasingly look like a terrible deal to hire 10 people for $80k/year when you could hire 2 people for $400k/year. And we are starting, culturally, to feel this vibe. In Quebec, following an IT catastrophe, it was recommended that the government create a team of top-gun tech people. It is left unsaid that these people will have high compensation. The evidence is increasingly undeniable that vast teams of regular people cannot match a few high-agency tech masters with AI. This is nothing new in the Silicon Valley... but it will increasingly hit closer to home.

We just went from horse-and-carriage to car. The old manuals still have truth in them, but the whole map of effort, speed, and leverage changed overnight. Building used to mean holding an entire system in your head. A fragile memory palace. A house of cards in biological primate RAM. If you stop, the palace collapses. Dinner, a meeting, a context switch. You come back and it’s glass dust on the floor. That’s why builders can look “antisocial.” It’s not vibes. It’s survival. You’re trying to get the palace out of your head and into code before it evaporates. Then the weird miracle: I type a few paragraphs that barely make sense on reread, and the machine builds the palace anyway. It mirrors the structure. It fills the gaps. It hands it back. The feeling is not “wow productivity.” The feeling is: I am seen. Like the part of you that has been translating yourself for 20 years finally gets understood on first contact. This is why the new skill is not “code faster.” It’s taste, direction, and leadership. Managing a swarm of agents. Running tight loops. Knowing what to ask for. And it brings back something old-school: apprenticeship. We forgot how to teach. Now teaching matters again, because the tools are insane but the mind behind them still has to be trained.

Here is my latest iteration of my /plan-exit-review skill for Claude Code. I use this as I exit plan mode and it works super well to shake out all issues,, shake out architecture and code smell issues, perf issues, and finally make sure every part of a PR is tested (I avg 1.3 lines of test to 1 line of real code these days) gist.github.com/garrytan/001f9…