How does the Claude Code team build Claude code? Today's deepdive covers this. An interesting take from them: "Mockups feel like a thing of the past when you can build 5-10 working prototypes per day." The full article: newsletter.pragmaticengineer.com/p/how-claude...

The deepdive is out: newsletter.pragmaticengineer.com/p/how-claude...

Ironic: Gartner had my post ridiculing Gartner removed. The one that shared Gartner's Magic Quadrant about AI Code Assistants, which does not include Claude Code, OpenAI Codex, and ranks Amazon above Cursor. This one: github.blog/ai-and-ml/gi... Are they embarrassed to share their Magic Quadrant?

Amusing that if you have yet to hear about Claude Code and work in tech: you are either living under a rock, or are someone who keeps up with the latest and greatest with an incredibly expensive Gartner subscription (or thinks they are keeping up, that is)

Until now, programming was a "single-threaded" activity. As a dev, I'd get "in the zone" and get it done. AI agents change this: you can now kick off parallel coding tasks. I see more devs do this... and this feels like brand new territory. We'll need to learn. A lot!

Companies that hiring a lot more interns vs before: - Shopify (1,000 interns this year) - GitHub (significantly upping intern hiring) - Now, Cloudflare (1,111 interns - a nod to 1.1.1.1!) Shopify + GitHub told me they do it in-part because interns bring lots of fresh ideas in how they use AI

Someone asked me about my advice to devs for "traditional software engineers" who resist change. My advice: Being a true tech professional means you keep your toolset sharp, and up-to-date. If you refuse to do so, you are probably not a true professional, but (cont'd)

One interesting outcome of AI agents being so fast at prototyping: It can take longer to have a meeting about building a prototype than to build it! So... why not just build it, then show it and talk about it?

What a hot mess. Official US social media accounts are now contradicting the published predidential order, saying it doesn't apply to existing H1B holders (the execive order says it does) and existing H1Bs would no be turned back on the border (the order says they would)

This book (Code Health Guardian by Artie Shevchenko) came recommended by an experienced software engineer as a great read. Starting it now. On first glance, it looks like a promising one, going back to software engineering basics that start to become more important with GenAI

As a reminder, in 2017, the first Trump admin already kicked the door on H1Bs by cancelling H1B premium processing. Many software engineers in the US had to leave the country. They went to Canadaian, UK and EU offices of US tech companies back then for a year. Also (cont’d)

Those on an H1B cannot return to the US from tomorrow (Sunday) unless paying $100K. This is an out-of-the blue presidential action. We’ll see software engineers stranded abroad. One easy to predict outcome: those on US visas will travel less… for work, for conferences etc.

I cannot describe how bad this is for tech companies. Not the $100K fee on *new* H1Bs, but that their software engineers currently abroad can be turned back (or detained?) at the border with ZERO notice! It brings uncertainty to running a business in the US. Plus (cont’d)

Talked with the Claude Code team on how they build Claude Code. It feels I get a peek into the future, and I get why Dario said 6 months ago that 90% of code will be written by AI. This team works SO differently than any eng team I saw. Will share in-depth soon. One example:

"Will AI replace software engineers?" Steve McConnell, author of Code Complete: "AI changes the playing field. Now you're trying to identify what is *exactly* right with AI-generated output. To me, software engineering is almost the same job it always was." Via The Pragmatic Engineer Podcast:

Conversation with a dev at a Well Known Tech Company Me: "What do you use for o11y?" Them: "We have our own, custom stack that the platform team built." Me: "Oh cool, must be nice." Them: "It sucks actually, the UX. Datadog is 10x better, but ofc we'll never go with that..."

I made a magic quadrant about how to choose your tech stack Seriously, DYOR: no one will do it for you. Try out stuff. See what works. Ignore reports from people who don't do the work and are pay-to-play anyway

One more reminder that Slack is a terrible choice for communities, and extorts even nonprofits. Even when it’s one teaching teens to code, for free. Hack Club is moving to Mattermost. They really have no choice. Just a poor taste how Slack demands $50K in a week, no real notice

All the good that TypeScript did to the Node ecosystem, npm is undoing it How can a serious company build on top of this ecosystem that is one npm update away from dependency compromise happening (I am serious: I'd expect more greenfield startups to not go with Node b/c of it)

In 2022, the US banned high-end NVIDIA GPU exports to China, wanting to slow down AI development in China. So China invested in homegrown AI chip production, and caught up in 3 years. Now they are banning buying NVIDIA AI chips. The kind of secondary effect a ban can have:

Interesting trend: Startups are getting rid of takehome exercises during interviews. Hearing it from @alinelerner.bsky.social The reason is - as you can guess! - AI. Most candidates just input the takehome to eg Claude Code, wait, and submit the result. Not getting much signal...

I keep observing how a bunch of large companies have *really* smart engineers doing *really* complex+challenging+interesting work and... no one knows about it outside. I only get to know about it by chance. Sometimes it's the PR policies stopping talk about this, but more often:

Worth remembering how 100 years ago, renowned electrical engineer Charles Steinmetz forecasted: within 100 years, industrial use of electricity would enable a four-hour workday. He was 100% right about the impact of electricity. And yet, 100 years later we still work 8+ hrs per day…

This is really interesting research. Details from 18 well-known tech companies, about how they attempt to measure + answer “what impact do our AI tools have on software development?” Excellent research by @lauratacho.com, details shared here for the first time:

One of the most common uses of "vibe coding" I'm hearing from professional devs, outside of prototyping: Building internal-only tools! Where you don't need to worry about security, scalability, malicious usage. E.g. data visualization / data viewer tools. Used a lot for this!

No idea why or how, but a doc I created as an example for a "work log" for software engineers a few years ago - plus linked in The Software Engineer's Guidebook - is getting a ton of activity (as in 5-10 ppl having it open regularly, anytime I check). This one: docs.google.com/document/d/1...

Controversial advice: rewrite software three times to get to standout quality. An interesting observation by Steve McConnell, author of Code Complete. Full episode: • YouTube: www.youtube.com/watch?v=iPKm... • Spotify: open.spotify.com/episode/2klD... • Apple: podcasts.apple.com/us/podcast/c...

Interesting: to buy a Playdate console (great device!) outside the US, one needs to pay US Tariffs… because the Playdate ships from Malaysia to the US, and then internationally. Because they are a small business and have no warehouse outside the US. Appreciate the transparency tho

So much talk about how AI coding tools can be used to generate software from scratch Yet when I talk with professional devs, the single best use case is... migrations! (and it's very useful there, saves a bunch of time and devs are happy to use it)

Code Complete is one of the most enduring books on software engineering. Steve McConnell wrote the 900-page handbook just five years into his career - decades later, Code Complete is still a best-seller. Steve rarely gives interviews, so I hope you enjoy this special one.

We're researching how software engineers are using MCP servers - interesting use cases, learnings so far, best practices. If you're either using MCP servers or building them, we'd love to hear from you! (Hoping to also highlight interesting uses - assuming you're ok with it!) DMs for Elin open:

More than half of of startups announcing how they hit $100M ARR the last few months I never heard of before, nor do I have any idea of what they do I feel there’s an “ARR overload” in tech: every second AI startup now sharing massive ARR numbers - and I stopped caring

I love seeing feedback on AI coding tools from companies which get feedback from their engineers, and evaluate tools based on this feedback + usage. Data from CloudKitchens: - GH Copilot: widely-used - Cursor: used - Windsurf: dropped - Devin: dropped - Claude Code: evaluating

Question from an intern at Uber: "If today you received an offer from a start-up vs Uber, which would you pick?" Early-career, I'd pick the option that opens more doors later on. If the startup is industry-wide known like OpenAI/Anthropic/Ramp: great choice. Else I'd go w Uber

For those saying "English is the new programming language." Option A: "Make it dark green ... no, lighter ... no, a bit less blueish ... no, a bit more pinkish ... no, a little more greyish ... no, a hint darker" Option B: use #7DB097 Which option is likely to stick?

Before AI coding tools: devs usually used the programming language they knew best (+ got the job done), OR spent a bunch of time and effort learning the language that they wanted/needed to use After AI coding tools: what will change? (How important are programming languages?)

With the Forward Deployed Engineer (FDE) role trending in discussions, here's our deepdive from a month ago in @pragmaticengineer.com on the origins of the role, how it looks at Palantir, OpenAI and Ramp, and why the role is so hot *now*: newsletter.pragmaticengineer.com/p/forward-de...

How can you become an advisor at a startup, as a software engineer (and get a healthy equity grant?) You need to be either very technical, or help close customers. From Peter Walker of Carta in The Pragmatic Engineer Podcast:

We should also acknowledge that for most AI startups that have not yet found product-market fit: they are racing agains the clock. Most probably have ~6 months (or less) to gain serious traction: else they will prob have to fold / sell It's much higher pressure than in the past.

Interesting observation: Amazon and Meta: very big on their core values/leadership principles (engineers working there reference them quite a lot, they are pretty relatable) Google's core values meanwhile... feel pretty abstract? Yet to hear a Googler reference them

For all the trends and talk about potentially fewer new grads hired in tech... ... when I think back of the best hires I made as a manager, these were: - 50% experienced folks knocking it out of the ballpark - 50% new grads who were SO motivated, smart, and heads down

One trend I hear across hiring managers in tech: They are looking for recruiters. Inbound applications are a such high noise / low signal for mid-level+ roles that demand for good recruiters to source candidates feels it's up, by a lot.

Today @suhailpatel.bsky.social told me something I still think about when we talked about measuring the efficiency of AI tools for devs: “Always applying critical thinking when using AI tools will be the difference between devs who do well with AI and devs who don’t” Resisting “autopilot!”

Amusing how 99% of people using LLMs forget how these things work: They are advanced probability machines. They generate the next most likely token (word) based in the input and their training. Under the hood, it’s a giant matrix multiplication that has eerily good output.

Interesting how Starlink’s biggest competitor is infrastructure. Got Starlink for a house w no connectivity. Worked great. But then cable got built out. Cable internet is faster, cheaper + lower latency, so switched Starlink keeps throwing deals at me… which are worthless now

If I was an engineer at Coinbase and heard this mandate, I’d just start to review AI code less rigorously and let it rip. The stated goal is to increase AI generation numbers, and nothing about quality, maintainability, correctness, lack of duplication - right? Give devs a goal and they meet it

Big news: Linear is the season sponsor for The Pragmatic Engineer Podcast! I’ll be honest: I never thought I’d be excited about a tool for project management or planning. All of the tools I used felt like they slowed me down. Linear was born to solve this frustration (cont'd)

Interesting observation from an eng manager: “As soon as I know some text is AI-generated: I lose all interest in reading it. For performance reviews, I asked people to either not use AI or if they must: just write down the prompt so I don’t need to go thru the generated word salad padding.”

Average tenure at Meta increased from ~2 years to ~4 years since the 2023 layoffs. Similar changes across most of Big Tech. Basically, folks working in Big Tech are not leaving like before. If that's not a smell on the market, not sure what is