AI on both sides of a PHP bug
A bug found with AI made universities and government agencies switch their websites off. AI also helped me check and patch my own sites: it prepared, I approved.
A bug found with AI made universities and government agencies switch their websites off. AI also helped me check and patch my own sites: it prepared, I approved.
Setting up the Hermes agent on a new Mac mini with Claude Code, one agent could read another agent's files. Isolating them was my job as an engineer, not the agent's.
This week I built a working ERP prototype in about an hour instead of writing a specification. When a prototype costs an hour, nobody defends it.
Vendors put AI into every seat and raise the price. With AI, a manager can build the prototype and a software team can finish it. The process should define the software, not the other way round.
I ran ten AI agents in parallel and was exhausted by Monday evening. Now I run one or two and switch context while they work. A marathon, not a sprint.
Meetings, Google Sheets and company data: what AI agents do for me every day. The agent prepares, I judge.
Five predictions on AI security: self-hosted models, a bigger role for security assurance engineers, and why the generating agent should not approve its own changes.
Where you put AI decides what you get back from it. Five slides on the two positions and what each one measures.
Microsoft put engineers on token budgets. The bigger story: for the first time in software, the cost side of a feature is metered like electricity.
Article 50 of the EU AI Act in force. Why I think disclosure is the right move, and why the scope covers more companies than most people assume.
Coding is getting cheaper. Knowing what to build is not. The programmer as sculptor, and as manager of AI.
Questions or comments: write to me at dmytro@durach.ai or on LinkedIn.