Logo
Join the CTO Newsletter for free!
 
Amazing CTO Logo

Amazing CTO | More happiness and success
🚀 111.4

by Stephan Schmidt

Happy 🌞 Sunday,

Welcome to my opinionated newsletter. This week’s insights

  • 🦹 How is an Engineering Manager at Meta different?
  • 🤖 Who codes better? Gemini 2.5 Pro vs. Claude 3.7 Sonnet
  • 💻 Is Junior the new Senior? The Revenge of the Junior Developer

Good reading, have a nice Sunday ❤️ and a great week,

Stephan
AI Evangelist and CTO-Coach

Need support as an engineering manager? Thought about coaching? Let's talk—I helped many CTOs and engineering leaders with growth and making the right decisions under pressure, I can help you too.
🎁

If you only read one thing

Being an Engineering Manager at Meta (24 minute read)

Many insights, “At Meta, managers do not manage a team. At Meta, managers support a team.” As it should be, but isn’t everywhere. “Software engineers are expected to choose what to work on. More senior engineers are expected to set the direction for their team and organization.” I think many companies can learn from that, but they only learn the things that are not useful, like using Kafka and Kubernetes. Sad.

https://zaidesanton.substack.com/p/being-an-engineering-manager-at-meta

🚀

Stories I’ve enjoyed this week

Gemini 2.5 Pro vs. Claude 3.7 Sonnet: Coding Comparison (19 minute read)

This is astonishing. I’ve already got excellent results with Claude Code, it seems that Gemini Pro is better (though my experiments with Cursor showed that Gemini misunderstands me more often - perhaps because I’ve worked more with Claude and it understands me better). Something to keep an eye out for, especially with my belief that AI is the new compiler, and we will move more towards one-shotting instead of continuous prompting. One would need a dev tool though.

https://composio.dev/blog/gemini-2-5-pro-vs-claude-3-7-sonnet-coding-comparison/


Revenge of the Junior Developer (45 minute read)

I thought this would be the one thing to read this week, but then it is a lot of predictions, with too little substance. Still worth a read though, many interesting thoughts, like that view of “Vibe coding is still completely invisible to 80% of the industry outside Silicon Valley” as if no world exists outside the valley. The author makes some things up, like agent clusters and agent fleets, and they conveniently start next year and then rise exponentially, but not today. I do agree with the code completions going down, but not so fast, as predicted (Non-existing in 2027). A very interesting point is how “Agent Fleets” are supposed to be double the amount of traditional programming (as mentioned before here, the limit is your ideas, and I don’t see more ideas coming, but hey, yes, when I started studying computer science in 1991 everyone told me all software was already written). Also chat-based programming will overtake traditional programming this year. I still do a combination of both. I still think the problem of vibe coding is cooperation. Five developers vibe coding on the same parts with five agents leads to problems. For now, all vibe coding that I see are single developers working on (side) projects. I don’t know what my agent does to the code, but when more than one developer vibe codes, how do I know what has changed? At one point, for that reason, we will need to abandon source code, but instead of prompting I think we will use AIs as a compiler for requirements and screenshots (and Figma) and database schemas into source code - that we won’t take a look at anymore (and new languages arising just for AIs and then in the end AI->binary). There is less substance in the rest of the article, but some funny things like “Just to make it concrete, you might tell a coding agent something like, “Here is JIRA ticket #; please go fix it.” Why would I need to do that? Can’t the AI just finish all JIRAs on its own? And then why JIRA, have a requirements.txt that you constantly change (You is the developer, or the CTO, or the PM, or the CEO, we don’t know yet) and pushes things to staging or to production, depending on your risk appetite. Those visionaries, trying to have a vision and then it stops right around the corner. Oh, but one good point left, the CFO will need to budget for those very expensive tools (unless we get, and I think we might, an AI-sympathetic local hardware), does your CFO? Why not, go, talk to them!

https://sourcegraph.com/blog/revenge-of-the-junior-developer


Feature Assurance vs. Quality Assurance. (6 minute read)

Great! Totally agree, developers do all the QA and PMs check if the feature meets the product view. If QA is doing the testing, you’re doing it wrong. I want to add that Quality Assistance is still helpful: QA people check and review developer tests (devs want them to run, to break them), do cross-feature explorative testing (because no one else does) and help PMs with a customer quality view when thinking about features. Again, if anyone else than devs write tests (except AI ;-) then you’re doing it wrong.

https://www.raphaelbauer.com/posts/feature-assurance-vs-quality-assurance/


The Frontend Treadmill (5 minute read)

Gold. “If you feel strongly about what framework you want to use, please make that a criteria for your job search. Please stop walking into teams and derailing everything by trying to convince them to switch from framework X to your framework of choice. It’s really annoying and tremendously costly.” Pure gold. “Companies that want to reduce the cost of their frontend tech becoming obsoleted so often should be looking to get back to fundamentals.” And on and on. Of course https://www.radicalsimpli.city/

https://polotek.net/posts/the-frontend-treadmill/


What to Do (10 minute read)

Another Paul Graham piece, in case you’ve missed it.

https://paulgraham.com/do.html


supabase-community/postgres-language-server (9 minute read)

NOW IS THE TIME TO PLAY WITH MCP, you are already, aren’t you? Or some of your developers are playing with MCP to connect an AI to your services and data and see what can be done? No? Then start with this easy Postgres MCP server. I mean it.

https://github.com/supabase-community/postgres-language-server


A Sneaky Phish Just Grabbed my Mailchimp Mailing List (21 minute read)

Troy Hunt, a (the?) security expert, has been phished with a fake Mailchimp email and lost his mailing list, all emails got leaked. Two insights when you read the article:

  1. Everyone can be phished, so be prepared
  2. Mailchimp should put dangerous actions behind email confirmation, e.g. you can only export all data with a confirmation email
  3. Do not have dangerous UI in your backend (yes, YOUR backend), like a list view of all email addresses that can be copied and pasted, that’s a feature everyone thinks should be there, but no one needs

(oh, that were three, well developers can’t count, only zero, one, many).

https://www.troyhunt.com/a-sneaky-phish-just-grabbed-my-mailchimp-mailing-list/


Hunyuan-T1 (10 minute read)

Another day, another new AI model. This time another Chinese one, from Tencent.

https://llm.hunyuan.tencent.com/#/blog/hy-t1?lang=en


Career Development: What It Really Means to be a Manager, Director, or VP (29 minute read)

My clients have often random titles, with no idea why they have them and what to expect. According to the article,

  1. Managers are paid to drive results with some support.
  2. Directors are paid to drive results with little or no supervision (“set and forget”)
  3. VPs are paid to make the plan.

https://kellblog.com/2015/03/08/career-development-what-it-really-means-to-be-a-manager-director-or-vp/


The Burnout Machine (8 minute read)

“It’s a meat grinder that chews up developers, sysadmins, and infosec pros and spits them out the other side - burnt out, disillusioned, and disposable.” You have the slightest feeling you’re burning out, stop reading, get help of a professional now. But the article goes much deeper, “We’re in an industry where burnout isn’t just common - it’s expected.” and “That’s the kicker. We’re the ones building the damn future, but we have no say in how it’s built.” As a capitalist, I think there is no way around unions.

https://unionize.fyi/


Moving away from US cloud services (14 minute read)

If you’re in the EU, should you move off US cloud services? I don’t know. But you should have a plan for your dependencies, if they go away, what you’re going to do and how long it takes. Just throwing the hands up isn’t professional, and it’s not enough, and I would be ashamed to take a $x00k salary and then throw my hands up. I had a client who was told by their payment provider they stopped serving them on very short notice. Do you have a second payment provider? If yours kicks you out? Or the country they are in decides that they can no longer serve you? And I sound like a broken record, do not have all your backups at the same cloud service!

https://martijnhols.nl/blog/moving-away-from-us-cloud-services


My AI Newsletter

Tabula - My AI Newsletter

I have another newsletter, about AI, with the focus on AI impact on engineering managers. Recently I wrote about the “AI’s Staircase of Control” which explains that the control you want to have over an AI is limiting the gains you get from it. Must read!

https://www.tabulamag.com/


Join the CTO newsletter!
Impressum