Insights

Lessons from driving business value with AI.

Everything here comes out of something we built, run or delivered for a client — including the parts that went wrong.

AI-first, humans in control: a deeper, faster, wider CMS assessment

How Levantar helped Highfield, one of the UK's top five awarding organisations, decide the future of its content platform — working AI-first, with people owning every judgement.

Read the piece

Letting an agent run code, without letting it run loose

Giving an agent a managed Python sandbox with AgentCore Code Interpreter, putting a file in, running code over it, and seeing what the session can and cannot reach.

Read the piece

The pruning has started. Most of it is cutting the wrong branch.

AI budgets are being cut on cost because value is the harder number. What the 2026 cost-governance data shows, and what to prune on instead.

Read the piece

Giving your agent memory that survives the session

Adding AgentCore Memory to an agent so it keeps conversation events within a session and recalls extracted user preferences across sessions.

Read the piece

An agent at the centre

Six best-of-breed apps that don't talk, four small open-source MCP servers, and an agent in the middle. What we built, what broke, and what it taught us.

Read the piece

Giving your agent tools with AgentCore Gateway

Putting AgentCore Gateway in front of an existing Lambda so an agent can call it as an MCP tool, authenticated from the first deployment.

Read the piece

How and why we built Upcast.Social

Why we built our own social tool rather than hiring or subscribing, how we validated it in Opptora first, and how it became a product with a reseller programme.

Read the piece

What AgentCore actually is, and getting a first agent running on it

How to deploy an agent container to Amazon Bedrock AgentCore Runtime with Terraform, invoke it, and see what the platform actually gives you.

Read the piece

Why we built Opptora first

We built our own validation platform so we would only build what we could prove had value. It now sits underneath the assessment we run for clients.

Read the piece

Claude + ElevenLabs → narrated MP4, in four languages

The narrated walkthrough, localised — Claude translates, ElevenLabs voices it in four languages, ffmpeg stitches each one.

Read the piece

Claude + Playwright + AWS Polly → narrated MP4

The same flow with video output: Claude, Playwright and AWS Polly producing a narrated MP4.

Read the piece

Claude + Playwright → interactive walkthrough

Point Claude at a feature and get an interactive walkthrough out the other end.

Read the piece

Building a multi-agent security pipeline with Google's A2A protocol and GitGuardian

Wiring multiple agents together with Google's Agent Development Kit and the A2A protocol, with GitGuardian scanning content for secrets along the way.

Read the piece

Contact

Talk to us before your next AI investment.

If something here matches a problem you have, the first conversation is diagnostic, not a pitch.