Notes on AI agents and MCP
Short, plain notes about building AI agents and MCP servers that work in real businesses: what works, what breaks, and why.
June 19, 2026
The code that writes the code
Anthropic now says more than 80% of the code merged into its own systems is written by Claude — up from low single digits before 2025. And this month it published a serious report on recursive self-improvement: AI helping build the next AI. Strip out the sci-fi and there's a practical message for the rest of us about where the bottleneck is going, and what discipline it demands.
- agents
- methodology
June 19, 2026
The government red-teams the model now
The US AI standards body signed deals with Google DeepMind, Microsoft, and xAI to evaluate their frontier models before public release — and has already run more than 40 assessments, some of models the public never saw. The UK signed parallel agreements. Strip away the politics and there's a clear signal: evaluation, not vibes, is how anyone actually knows what a model can do. Steal the pattern.
- security
- methodology
June 19, 2026
The AI speedup comes with a security bill
Gartner says 90% of engineering leaders report gains from AI coding tools — a net 19% productivity bump. The same research says unreviewed AI code has 23% higher bug density, and 14.3% of AI-generated snippets carry security vulnerabilities versus 9.1% for human-written. Almost nobody prints those two numbers in the same sentence. You should, because they're the same story.
- security
- methodology
June 15, 2026
It wasn't the model. It was your data.
Most AI projects fail — MIT found 95% of generative-AI pilots delivered no measurable profit, and RAND put the overall failure rate around 80%. When it goes wrong, the instinct is to blame the model: not smart enough, wrong choice, bad prompts. The data says otherwise. The single most-cited cause of failure is poor data quality, and only about 12% of organizations have data clean enough to support AI at all. You probably don't have a model problem. You have a data problem wearing a model problem's clothes. Here's how to tell.
- architecture
- methodology
June 15, 2026
Resolved — but they wanted a human
Companies love the number: our AI resolves 76% of support tickets on its own. Customers are telling a different story. Across 2026, the share of people who'd rather talk to a real person rose to 85%, frustration with AI agents climbed to 59%, and more than half will abandon even a solved AI-only chat if the path to a human feels blocked. 'Resolved by the bot' and 'happy customer' are not the same thing. Here's the metric you're probably missing, and how to stop optimizing your way into a backlash.
- business
- methodology
June 15, 2026
Spending up, confidence down
Companies are pouring money into AI — budgets up sharply, some doubling year over year. And in the same breath, 51% of CIOs say adoption is already moving too fast for them to manage. That's a strange combination: the people writing the checks think the thing they're funding is outrunning them. The reflex is to read that as 'slow down.' The data says the opposite. The teams that move fastest aren't the cautious ones — they're the ones who built the guardrails first. Here's the real lesson hiding in the contradiction.
- business
- methodology