My personal anecdote on the impact of AI (aka Claude Code + Cursor)

There was this tool I wanted to build that would have helped my business a little bit at @Pragmatic_Eng, but not enough to

1. Do it myself (would have taken ~days)
2. Hire someone (too much onboarding)

But... ... but then I gave it a go with Claude Code + Cursor open.

In 30 minutes, I had something promising. In an hour, I got it done, exactly how I wanted (this was an addition to my existing codebase.) In another hour, I moved it over to a new stack I've been wanting to play with!

Linear has a 30-minute weekly meeting called "Quality Wednesdays." I sat through one and WOW

Devs show a quality or perf-related fix they did last week. It can be big, or small. We went through 17 issues, from massive backend performance wins, to this tiny one. Can you see it? On one end, it was super casual. On the other, it was really dense.

A dev spent 2 minutes showing how because styled-components *feels* slow, they tried out 3 other frameworks & measured how they compare for build time, JS bundle size, CSS size, and LCP rendering performance.

Based on this, they'll probably move off styled components... mainly for LCP rendering for massive pages to be faster. But it's all tradeoffs.

Amusing use of LLMs at a more traditional company:

“A project with ~50 people got stuck. There are too many JIRA tickets, no clear specification, and anytime one team tries to make progress, the others shoot it down.

So a dev built an LLM to try and break the deadlock: (cont’d) - Fed all JIRA tickets to the LLM. Built a basic RAG with vector DB

- Had it generate questions about the project, about topics not covered by the tickets

- Had the LLM attempt to answer the same questions

- Generated a report of what areas are not specified

- Tried to use this to stop teams rejecting suggestions “because this is not well specified”

Regarding the Windsurf sale (part of the team acquihired by Google, prob a great exit, but not all the team):

I feel we’re forgetting well-funded startups today are NOT scrappy startups in the past where employees work for pennies, paid well under market.

Its a different game What is true, and always has been true: founders and decision makers always have the biggest potential upside - for anything! Including negotiating and acquisition.

This is why so many accomplished employees eventually become founders - because its hoe you have more control of your destiny

There was this engineer on my team a while back who was: a good dev, but not the best dev. Got everything done. But had zero ego, a very nice personality, and got along with *everyone* on the team very well.

When he joined, the team became... better. Nicer. More balanced. I just got a reference check about this dev, asking the usual questions ("what is an example where they delivered over and beyond," "how did they execute", "what are growth areas" etc)

He did fine on all of them, but I still think how much better he made my team. With stuff that's hard/impossible to measure!

So predictable that we’ll see an explosion of digital products selling “ideas for million dollar businesses” that you can “just vibe code quickly”.

Basically: “buy my digital product for $500, spend $1,500 on Lovable / Claude Code and become a millionaire.”

Another hype train Ofc these products promoted by influencers will work just as well as crypto sh*tcoins launched by influencers in 2023.

We’ll see doctored evidence (“someone who built one of ideas idea is at $5K MRR after 2 weeks”) and nontechnical people will spend thousands for $0 in return

I generally like Anthropic: but the more they paint a dystopian future where AI “manages” people (“AI middle-managers”) the more I am starting to think they are losing their marbles.

LLMs is a tool humans should use. The tail should not wag the dog; Anthropic should know better

https://twitter.com/anthropicai/status/1938630310985601131

And frankly I’m getting tired of Anthropic being loud about how their AI will lead to mass unemployment, and while claiming to be a responsible lab to develop AI.

If your master plan is to wipe out the labor market for profit: you’re not responsible.

You cannot have both.

https://twitter.com/anthropicai/status/1938630312734474248

Something I hear very little talk about:

How AI coding tools are so much LESS useful when used on existing, large codebases at work (with custom frameworks, conventions, coding style etc)

... compared to doing greenfield work or side projects

So common for me to hear: "yeah I love it on my side projects, but at work it's 'meh'" I'm getting details talking with devs at the likes of eg Google, Meta, Microsoft: the companies building some of the best AI coding tools out there!

And yet, for their existing codebases, the usefulness is marginal. Mostly for autocomplete (that has a higher miss rate than for greenfield)

This blog is SO good at pointing out what should have been obvious about AI for coding (Copilot and others)

These tools are good for re-creating whatever they’ve been trained on.

They are not what will create the next, better generation of frameworks, libraries, technologies. Full blog - you should *absolutely* read it

I also find these AI tools helpful when it’s doing the routine task I’ve done many times and can do it with eyes closed

But… it’s not helpful when I want to build something GREAT that is elegant, and better than beforedeplet.ing/the-copilot-de…

I am hearing SO many stories about people realizing coding with AI tools (aka “vibe coding”) is a game changer after “reviving” an old side project or idea on the side and making so much progress

But… while I often hear the excitement on starting: not hearing “finished” often! Almost like these tools were amazing at making rapid progress at first… but it still takes a ton of effort to finish things and feels like most people go back to leaving side projects unfinished (even if in a more advanced state?)

Question from an ex-Uber engineer:

"I got this reachout from recruiting Uber. I responded that I'm happy to discuss why I left (so Uber can learn from it) but not planning to return.

I got ghosted. Why? They asked, after all!"

Here is exactly why (continued): It's b/c you mis-read the email (which is so easy to do!)

It sounds like a "we'd love feedback and improve", right?

WRONG

This is a recruitment email, using Uber alumni as a high conversion channel.

It's from a sourcer: who only has one goal: get ppl in the hiring pipeline!

Can we just mention what practically everyone using Cursor and Windsurf uses/pays for as version control again

And who owns that service

https://twitter.com/rhyssullivan/status/1919609657523310785

In case you missed it: it was Microsoft who voluntarily cannibalized their very very profitable Visual Studio business and released VS Code for free. And made it trivial to fork. VS Code + forks probably account for 80%+ of the global dev market in usage

Why did they do it?

I’m tired of hearing the “AI is killing tech jobs” narrative. Here is data on “top” tech companies and startups hiring.

The last 2 years (since GenAI went mainstream and AI coding tools evolved greatly) we’re seeing more hiring from them. Below the pandemic peak ofc Data source: @Pragmatic_Eng

newsletter.pragmaticengineer.com/p/the-reality-…

And before you show me the tech jobs going down graph that goes viral every week: know that most sectors see the “decline in jobs” from the pandemic peak: blog.pragmaticengineer.com/software-engin…

Here’s one reason Apple fought tooth and nail to disallow web payments for apps:

Because Apple’s IAP is bad in many ways, and *so many* apps will move to web-based payments now not mainly because of the 30% Apple fee, but because of how bad IAP is.

Let me give you examples: 1. Refunds

With Apple IAP it’s just not possible to do!! No, it really is not for the merchant. They cannot do a full or partial refund. Talk about poor customer support!

2. Group subscriptions. Nonexistent with IAP.

3. Paying using a non-credit card option. IAP does not allow

Every now and then there's this prediction of when we will see the first one billion dollar company ran by one person...

... and I think back to how in 2016 there was this one product inside Uber that had crossed a $1B annual run rate that had a total of one dev allocated to it. And half a data scientist (part-time).

It was cash.

Funny how headcount games can work inside fast-growing companies, especially when the product is a stated goal of what a founder does NOT want to support (but turns out to be essential!)

From an eng manager at a full remote company:

"We just fired an engineer after ~15 days on the job who lacked basics skills on the job but aced the interview - clearly, using cheat tools.

He admitted to how he did it: he used iAsk, ChatGPT and Interview Coder throughout" (I personally talked with this person and know them well)

This company hired full remote without issue for years: this is the first proper shocker they have.

They are changing their process, of course. In-person interviews, in-part likely to be unavoidable.

I am coming around to why MCP is so impressive.

For one of my side projects, I used to have to log onto my database admin (PgAdmin) to query stuff.

I connected an MCP server to Postgres and can "talk" with my database (and data!)

An uplevel in my productivity + ease of work. Let me give an example of how cool this is:

I have a service that sends our promo codes for Perplexity and Kagi that paid newsletter members can claim:

I can now "talk" with the data from my IDE to get all kinds of details!

Was impossible until now newsletter.pragmaticengineer.com/p/free-kagi-an…

I'm starting to understand why there are company eng blogs not worth reading.

When doing a deepdive on an interesting company in @Pragmatic_Eng, we do research, talk with engineers, then share the draft back for any minor corrections. Usually it's a "LGTM." But sometimes: Sometimes the Comms or Brand team gets actively involved, and mistakenly assume they are the editors, and attempt to rewrite the whole thing on how they would usually publish it on eg their blog.

Every time, it's a disaster to see, but also amusing. Because a good article becomes SO bad. Interesting details removed, branding elements added etc.

(We never allow edits - and if they insist we simply publish nothing, throwing out our research. This has not yet happened, but it might be the first time it will)

One thing that really bugs me about VCs and others projects claiming how AI will mean many devs redundant because smaller teams can do more with less: is ignoring the last.

Some of the most impactful / successful software was built by tiny teams in the 80s, 90s, 2000s. Like: Microsoft’s first product in 1975 years ago: 2 devs

Quake in 1996: 9 devs

Google’s first search engine in 1998: 4 devs

We could go on.

Small teams with outstanding people doing great things happened before GenAI and will happen after as well (and without as well!)

A good reminder why you can pick up GenAI - and you probably should. Real story:

Small company, 5 devs. Last time they hired was 12 years ago. AI comes out: company wants to add AI feature. But they don't have the expertise. So hire an AI agency.

Agency spend 3 months planning: After 3 months, the present a very complex architecture to build: several services multiple databases, SageMaker models etc, using a language a company is not using (Python - this is a Java shop)

It will take 6-9 months to build

Operational costs will be higher fort this one feature than all of the SaaS operational costs for the company!

Klarna was the company that went all-on replacing customer support with an AI bot and went on to brag about the cost savings.

Now they are reversing course.

Easy to see more companies blindly replacing quality customer support with a worse AI implementation will follow... Back when Twitter was full of influencers declaring the end of customer support thanks to Klarna I did something few people did:

Signed up for Klarna, bought an item, and used the bot.

I was NOT impressed. At all.

Called that this was... very basic. blog.pragmaticengineer.com/klarnas-ai-cha…