Gergely Orosz

Charlie: The teammate every TypeScript team is missing. He reviews code, fixes bugs, opens PRs—directly in existing GitHub and Linear workflows.

Here are three you probably have not heard of (no affiliation w them, just proving the point)

Charlie Labs "the teammate for TypeScript teams" charlielabs.ai

Fine Dev "build and launch anything in minutes" fine.dev

KAVIA AI - logo looks awfully like Claude kavia.ai

Incredible how many vibe coding tool startups are launching left and right.

It's become so trivial to build one of these (you can do it in a few hundred lines of code on top of a cutting-edge LLM like Claude/Gemini/ChatGPT) that it's no wonder.

It's becoming a marketing game

One I have been recommended and am reading: Software Architecture by Mary Shaw and David Garlan. Published in 1996 (good doesn’t mean it has to be new!)

(Thanks for this specific recommendation @booch.com)

What is a book you learned a lot from related to software engineering? (Aka one you'd recommend)

Comparing Heroku to other vendors will never be fully fair. But it's fair to compare Heroku to their old self.

Heroku in 2010 >> Heroku in 2025

In terms of:
- The urgency to handle outages
- TTD and TTM
- How serious they took follow-up steps

newsletter.pragmaticengineer.com/p/why-reliab...

yes, it's my takeaway as well, looking at how Heroku handled outages in 2010 vs in 2025.

2010 Heroku >>> 2025 Heroku in how they handled incidents, and how serious they took follow-up work

The biggest question is why this outage (all of Heroku down!) was not an "all hands on deck" situation. Why were there no parallel workstreams to get the service back up online kicked off?

Or if there were: why does the RCA not capture this?

From newsletter.pragmaticengineer.com/p/why-reliab...

Given the Heroku team's first action was to engage with a "third party vendor" to debug networking issues, it's fair to ask:

Has Heroku outsourced its technology expertise? If so, why?? As a cloud provider, this detail is the weirdest of all.

Especially that the issue was in-house...

Why reliability is hard at scale: learnings from infrastructure outages What are the lessons of large outages at Heroku, Google Cloud, and Neon? Plus, how responses to outages can be as informative as incidents themselves…

More analysis on the inexplainably poor handling of this outage by Heroku (e.g. 8 hours to post a status update to X!) newsletter.pragmaticengineer.com/p/why-reliab...

The RCA that brings up a lot more questions: help.salesforce.com/s/articleVie...

Heroku's infra team has some serious soul-searching to do.

On 9 June 2024, an Ubuntu 22.04 update caused outages at eg OpenAI, GitLab and others. Those outages lasted 2-6 hours. Heroku's was 23 hours.

So so many questions on Heroku stumbling to resolve the outage. Eg. why delegate to a 3rd party?

The Pragmatic Engineer 2025 Survey: What’s in your tech stack? Which tools do software engineers use for backend development, frontend, infrastructure, AI tooling, and more, today? Reader survey, with feedback and analysis, based on 3,000+ responses

GitHub Copilot is the most frequently mentioned AI tool by software devs as per The Pragmatic Engineer 2025 survey:

newsletter.pragmaticengineer.com/p/the-pragma...

My deepdive with Thomas about lGitHub in this recent podcast episode: newsletter.pragmaticengineer.com/p/github

This number (20 million+ users using GH Copilot) is wild because… most estimates put the number of all professional software developers in the world somewhere between 20-30 million.

GitHub Copilot either is getting ~100% adoption (unlikely) or it’s expanding the dev pool (likely!)

Migrations Done Well A guide for executing migrations well, at both small and large scales.

An educated guess is this is a migration gone terribly wrong… again, thanks to assuming they can do a sloppy migration with long downtime.

The way you do migrations well is planning for zero downtime. Or the very least have a robust rollback plan. More: newsletter.pragmaticengineer.com/p/migrations

When preparing for a zero-uptime upgrade you invest a LOT of time into things like shadowing, monitoring, having a rollback plan that works etc.

You do none of this when knowing “we’ll have 6 hours to take the full site offline.”

So you don’t know how to even rollback. Embarrassing, honestly

I wonder if old-school companies used to doing updates by going offline for hours cause more outages thanks to refusing to plan zero downtime updates.

Dutch mobile carrier Odido (formerly: T-Mobile) scheduled a site+app update for 6 hours. It’s down for 32 hours and counting…

Yes! They built this for 2-3 years early on, and was THE differentiator.

It’s what allows so seamless multi-editing / multi-work and why no other competitor compares…

Built king of like a game engine!

Everyone surprised Figma uses Ruby might be missing the forest from the tree.

What is amazing about Figma’s tech stack in my view is building a full-on cross-platform rendering engine in C++ and how web devs still wrote C++ at Figma (!!)

More here: newsletter.pragmaticengineer.com/p/building-f...

Figma is going public today (31 July). Here's their tech stack, at a high-level, from the deepdive with CTO Kris Rasmussen in The Pragmatic Engineer back in 2023:

The full deepdive on how Figma operates from an engineering point of view: newsletter.pragmaticengineer.com/p/inside-fig...

One of the most surprising learning for me is that LLMs are *not* a good fit for generating + maintaining SDKs (from eg one language to another) due to their non-deterministic nature!

BUT can be a good fit to help build the automated tooling that generates the SDKs from a spec:

The reality of offering “unlimited” plans for any service where the provider has costs based on usage.

There will always be users who abuse it in ways the provider doesn’t expect it (at least at first).

For AI, generating tokens costs $$. So “unlimited” is either subsidized or removed:

YouTube video by The Pragmatic Engineer Amazon, Google and Vibe Coding with Steve Yegge

Check out the full episode:

• YouTube: youtu.be/TZE33qMYwsc

• Spotify: open.spotify.com/episode/58fY...

• Apple: podcasts.apple.com/us/podcast/a...

• Summary and transcript: newsletter.pragmaticengineer.com/p/amazon-goo...

What a story from Steve Yegge from his time at Google, back in the day:

Using a "blind" and unbiased process, interviewers would not have hired most colleagues who were actually at the company!

Food for thought on how building a soild hiring process remains tough

My take:

1. This place sounds as dysfunctional as most similar places - the project will eventually get done

2. Honestly good on the dev working there who took initiative (not that common at such places!) and got hands-on experience with LLMs, doing a cool experiement! (the thing took a few days)

Not from this example! Those on the project saw the problem but it’s an organisational issue not a tool one!

From all I understood the project suffers from no clear project lead, and teams looking out for their own interest and no one taking the initiative to step in as lead (“not my job”)

AI didn’t solve for this. But it didn’t make things worse and created an opportunity to experiment with it.

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

A PM at this company told me this story. Asked him if this LLM helped break the deadlock? His response:

“No. We’re still stuck. But it was good excuse to play around with vector databases!”

Ha.

- 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

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)

YouTube video by The Pragmatic Engineer Measuring the impact of AI on software engineering – with Laura Tacho

The longer conversation on measuring the impact of AI on software engineering:

• YouTube: www.youtube.com/watch?v=xHHl...

• Spotify: open.spotify.com/episode/4hDo...

• Apple: podcasts.apple.com/us/podcast/m...

• Summary and transcript: newsletter.pragmaticengineer.com/p/measuring-...

"One of my unconventional opinions: Roadmaps are on the way out in the age of AI."

- by @lauratacho.com, CTO at DX: