K
@KevinTrethewey
Followers
1K
Following
4K
Media
503
Statuses
9K
Professionally interested in systems and people that produce software. Amateurishly interested in everything else. Head of Eng & Platforms; https://t.co/e0yLMvl41c.
Johannesburg, South Africa
Joined November 2007
After 25+ years in software, I've seen my share of hype cycles. But agentic AI is different. It’s not just faster tools—it’s a shift in what software even is. 🧵 A guide for veteran engineering leaders on how to recalibrate:
3
0
59
⚡️What she is feeling is skill deflation shock. And it is real. But the conclusion most people jump to is wrong. 1. The uncomfortable truth The pain is not that AI learned her craft. The pain is that her craft was procedural mastery, not scarce judgment. High-end
The fact that it took me years to master high end retouching and frequency separation in photoshop, to the point where I developed fcking carpal tunnel before the age of 25 now people can just ask chat to do it edit anything for them is something that I have a hard time
75
151
1K
Last quarter I rolled out Microsoft Copilot to 4,000 employees. $30 per seat per month. $1.4 million annually. I called it "digital transformation." The board loved that phrase. They approved it in eleven minutes. No one asked what it would actually do. Including me. I
5K
24K
163K
The bitter lesson for software engineers is that, in an economic system that prioritises shareholders over customers and employees, power and perception matter more than competence and reality.
0
0
2
We do not need universal basic income. We need universal basic meaning. Cash without purpose still breaks people. Purpose without cash is cruelty. What would a meaning first society give every citizen by default? Reply with one idea.
509
44
576
I went on a job interview for a Senior SWE role. They asked me about my experience with Kafka. I told them how in “Metamorphosis” someone can lose their worth in others’ eyes the moment they stop being useful. It’s a chilling reminder of how fragile our sense of belonging can
160
474
6K
Pick your poison: https://t.co/tj1mnI3llo OR https://t.co/bFO4OqWQgS
The way we treat AI affects the results. Being polite gets better results. Being impolite produces low performance and bad results. "Our study finds that the politeness of prompts can significantly affect LLM performance. This phenomenon is thought to reflect human social
0
0
0
My top learning for 2026: If you want to understand and predict a human system (workplace, community, political system etc), don't only look at it's goals and purpose, look at its incentives and constraints.
0
0
1
And - it only changes faster from here, and now it's coming for all careers. Us software folks probably have some useful coping skills that are going to need teaching...
0
0
0
I was born into a home with no personal computer, no internet, no cellphones. Not unusual for the late 70s. In my career lifespan in software I've seen my field go from "four characters is too many to store the year" to "Run these 10 documents through an AI on the other side of
2
0
0
All the great breakthroughs in science are, at their core, compression. They take a complex mess of observations and say, "it's all just this simple rule". Symbolic compression, specifically. Because the rule is always symbolic -- usually expressed as mathematical equations. If
494
748
6K
Why do experts often not make great mentors or teachers? Surely if they understand it, they should be able to teach it? Can't they just decide to share their wisdom? Nope. They can't. If you do something often enough, at some point you develop enough intuition that relying on
0
0
0
HigherGround, a legit South African non-profit, is raising funds for their therapeutic adventure holiday for people with disabilities - taking place in December ( https://t.co/fkybuf1NqR). Consider making a small donation - lots of small contributions will add up to massive
goodthingsguy.com
Higher Ground runs therapeutic holidays designed for people with disabilities. They are planning a December programme for blind people.
0
0
0
The digital intelligence paradox: The more generally intelligent the AI the more context and expertise required to use it. It’s the “paradox of choice” in digital form; the smarter the system, the more you must understand what you _actually_ want.
0
0
0
This aligns strongly with my experience building an agentic learning loop based system. Some helpful principles I’ve isolated so far: / Focus on agency and learning. Intelligence is an input to those. / Reason about digital intelligence and analog/real world/human/tacit
kevintrethewey.com
My Personal Site This site is a collection of personal writing and projects, including sporadic writing, and hobby projects.
One of the really fun parts of building AI agents is iterating to figure out when something should be handled by the main agent, should be a deterministic tool call, or be a subagent. * Main agent - great for keeping context and state of the overall workflow, but eventually you
0
0
1
The bottleneck for deep skill isn't usually intelligence, but boredom tolerance. Learning has an activation energy: below a certain skill threshold, practice is tedious, but above it, it becomes a self-sustaining flow state. The entire battle is persisting until that transition.
122
906
7K
Vibe coding gives you a running solution, not a working solution. The gap between those cannot be closed using vibe coding. That doesn't make vibe coding wrong or ill advised, but you are walking through a minefield (or sending others over one) if you don't come to terms with
0
0
1
Peak irony: @Grokipedia does not have a page for "vibe coding". (posted with the self awareness that I am bashing a v0.1 solution that will get better very quickly (but using it right now very much feels like I'm using something someone vibe coded))
0
0
0