Prompt injection: limit what a fooled agent can do
No filter today reliably stops an AI agent being fooled by instructions hidden in what it reads. What you can control is how much a fooled agent is able to do.
· 8 min read
Skip to content
Blog
Practical notes onsoftware and AI
What we learn building AI products, modernizing older systems and running software in production. Written for the people who make the technical decisions.
Prompt injection: limit what a fooled agent can do
No filter today reliably stops an AI agent being fooled by instructions hidden in what it reads. What you can control is how much a fooled agent is able to do.
8 min read
MCP explained: connecting AI agents to your systems
The Model Context Protocol gives AI agents one standard way to reach your tools and data. What an agent may do once it’s connected is still your decision.
8 min read
Cyber Resilience Act: what the new reporting duty asks
Since September 11, 2026, makers of software products and connected devices on the EU market have had to report actively exploited vulnerabilities and severe incidents, starting with a warning within 24 hours. Here’s what the duty asks, who it covers and what to have ready.
8 min read
Do AI coding assistants make teams faster?
Developers often feel faster with AI coding tools. Measured results are less certain, and the team’s delivery system decides much of what the tools are worth.
7 min read
How to test an AI feature before it launches
A demo shows that an AI feature can work. Before launch, you need evidence of how often it does, on the cases your users will bring.
8 min read
Internal developer platforms: what to build first
Nine in ten organizations in DORA’s 2025 research use an internal developer platform. DORA’s advice on what to build first: one journey developers repeat often, made clearly better, with speed and stability measured together.
8 min read
The strangler fig pattern: modernize without a rewrite
Replace an old system one piece at a time while it keeps running. Moving the first piece is the easy part. Switching the old one off is the milestone.
9 min read
Data contracts: how to stop downstream breakages
Write down what a shared dataset promises, and check it where changes are made. Writing the contract is the easy part. The milestone is a breaking change that fails in the producing team’s pipeline, before anyone downstream sees it.
9 min read
How AI agents can transform your business website
Most websites still work like brochures: they wait. An agent turns yours into a team member that answers, qualifies and completes work the moment a visitor needs it.
3 min read
No articles match that search.
Working on something similar?
Tell us what you’re building or fixing, and we’ll share how we’d approach it.