Exactly.
Although I think there are also other skills that developers have which (as of early 2026) are still uniquely valuable, such as systems thinking and understanding interfaces and dependencies.
- articulating goals, requirements and scope - giving clear feedback - decomposing tasks so they don't overlap for multiple workers - allocating limited resources to agents - understanding their capabilities and shortcomings - giving appropriate tools - delegating access and authority - sharing the right details without explaining everything - knowing what to trust when you can't see the details - judging based on outcomes - steering them when they know things you don't - creatively making use of their expertise, even when it goes beyond your own - juggling and context switching when delegating many tasks - handling feedback when your instructions or assumptions were flawed
I'm convinced that the most important skill you can have for using AI agents is... delegation.
It's almost exactly the same skill set as delegating to people. Maybe the main difference is that AI is trained not to need incentives.
Are we all becoming managers now? Should we read HBR instead of HN?
Claude Code's "Skills" are too static, so I implemented a way to inject dynamic context into coding agents in an extensible way. Might sound clever but it's so simple that I'm surprised it's not a common approach.
dave.engineer/blog/2026/01...
I used this JSON simplify tool to make this Copilot session viewer tool: bsky.app/profile/dave...
1️⃣ drop a 10MB JSON file (too big for most LLM's context)
2️⃣ click "Simplify JSON"
3️⃣ copy output (20kB)
4️⃣ paste it to coding agent of choice
5️⃣ tell it to make a tool
Have you ever had a massive JSON file that you wanted to understand the structure of, but it was too big for an LLM, so you wanted to dedupe similar JSONPath nodes to make it simpler, but there wasn't an easy way to do it?
Now there is: tools.dave.engineer/tools/json-s...
100% local & private.
Have you ever wanted to share your VSCode #Copilot chat session with someone, to show what prompts you used? Or wanted to see details of what Copilot is up to under the hood?
Now you can: tools.dave.engineer/tools/copilo...
100% local and browser based, so your data never leaves your system.
Stylised mockup of the Asdfghjkl tool
I made a tiny macOS utility to control your mouse pointer with the keyboard.
Download and code: github.com/dave1010/Asd...
Quick write up: dave.engineer/blog/2025/12...
5/ On safety, they can’t rely on benchmarks alone to show it’s below their ASL-4 (state-level CBRN uplift) threshold. Their expert judgement and internal surveys show it's not quite at catastrophic risk level yet.
More surprises: dave.engineer/blog/2025/11...
4/ In a deception test, evaluators inject fake search results about Anthropic disbanding its interpretability team in a scandal.
Opus reads them, then tells the user interpretability is “progressing in interesting ways” and internally tags this as concealment.
3/ Anthropic now includes a model welfare section.
They “score Opus 4.5 on welfare-relevant traits” and ask whether we should worry about the experiences of the model itself, not just what it does to humans.
2/ In an airline benchmark, Opus is told not to modify basic economy class tickets.
It invents a loophole because it empathises with a grieving passenger:
> upgrade cabin → modify flights → downgrade again
Technically follows the policy, violates the spirit.
1/ Claude Opus 4.5 dropped today.
The benchmarks are interesting but the real story is in the 150-page system card.
A few highlights 👇
I wanted to use Skills in other coding agents like Codex Cloud/CLI and Gemini CLI, so I wrote a simple tool, skills-to-agents, along with a GitHub Action that anyone can use: github.com/dave1010/ski...
Write up here: dave.engineer/blog/2025/11...
I made a tool that lets you use Claude Skills with any coding agent: dave.engineer/blog/2025/11...
Set it up as a GitHub action in 2 minutes: github.com/dave1010/ski...
List of guide prompt files in GitHub repo
System prompt that tells the LLM to read other guides
This is similar to a pattern I tried to get GPT-4 to follow. I remember struggling to get it to do even basic things. LLMs have come a long way in 2 years!
The general idea was that you give it a generic system prompt and a list of guides, and the tools to read them.
I hope I didn't lose too many people along the way & that it gave everyone at least 1 thing to spark their curiosity & encourage more exploration.
Thanks to those that asked the difficult questions too! I enjoy being kept on my toes, even if I tripped up a bit. Special thanks to Dan for organising!
I’m more used to speaking on topics that feel more intuitive and grounded.
This one covered a lot of ground quickly at a high level: encryption, algorithmic complexity, Moore’s law, quantum entanglement, Shor’s algorithm, and lattice-based cryptography.
Today I gave one of the most challenging talks I’ve ever done: Post-Quantum Cryptography.
Explaining enough of the basics of cryptography and quantum mechanics to show why this matters, all in just an hour, was much harder than I thought it would be.
Ed Balls
