The blog
Our take on scaling AI in the enterprise: use cases, integration, security and adoption.
Our approach to AI in production →
- 6 min
AI Platform Engineering: exploration, capitalisation, industrialisation
A free playground, shared capitalisation, strict industrialisation: the AI Platform Engineering method for taking AI agents from prototype to production.
Read the article - 7 min
MCP architecture in a regulated environment
Exposing many MCP servers to an agent without widening the attack surface: read-only by default, strict domain separation, tightly scoped permissions.
Read the article - 7 min
Evaluating an AI agent before production
A successful demo proves nothing. The criteria of an evaluation framework (determinism, reversibility, observability, consensus) for a reliable AI agent.
Read the article - 6 min
FinOps for agentic AI: measuring an agent's real cost
Cost per run, tokens, Wh, gCO2e: how we apply FinOps to agentic AI to track, attribute and cap the cost of AI agents just like a managed cloud budget line.
Read the article - 6 min
Why you need an agent registry in production
Past three or four AI agents in production, an agent registry becomes necessary: traceability, versioning, reuse and governance across the whole fleet.
Read the article - 6 min
Choosing between Claude models by use case
Route each task to the right Claude model by need: fast triage, everyday task or long reasoning, rather than trusting a single benchmark to decide for you.
Read the article - 6 min
Amazon Bedrock AgentCore: taking your AI agents to production
An agent that works in a demo isn't an agent in production. AgentCore brings the missing runtime, memory, identity, tools and observability, here's how we use it.
Read the article - 7 min
How we built a Claude-powered NOC triage agent that learns from every incident it sees
We replaced an 11-minute manual NOC triage with a Claude-powered agent that investigates in under two minutes, grounds every recommendation in past incidents, and gets smarter each time.
Read the article - 8 min
Deploying Bedrock AgentCore under an IAM permissions boundary: the choreography no one talks about
Deploying Bedrock AgentCore in a regulated, boundary-constrained AWS account: eight IAM patterns — role naming, PassRole, KMS, Memory data plane, evaluations — the docs don't connect for you.
Read the article - 6 min
Cross-account access to Bedrock AgentCore: two patterns, four gotchas, and one critical bug
Two supported patterns for invoking a Bedrock AgentCore runtime from another account, the four IAM gotchas behind misleading denials, and the critical bug to avoid.
Read the article - 5 min
From demo to production: closing the gap that kills AI projects
Most AI projects stop at the demo. Here are the four obstacles between the idea and real usage, and how to clear them to create measurable value.
Read the article