Insights
Writing from the delivery floor
What we learned putting Claude into other companies' production systems — which workflows are worth automating, where adoption stalls, how guardrails get drawn, and what the migrations underneath all of it actually cost. Filter by topic or search below.
Every post
Claude adoption, agents and the systems under them
Newest first. The line here is the summary; the post is where the decisions, the constraints and the things that did not work get unpacked.
Why agentic AI needs more than one cloud
Agents want reach and ledgers want independence. Put both on one provider and you get a single point of failure for the AI and a single point of control for the record.
Read article →Claude on Bedrock, Vertex AI or Foundry: where should each agent run?
Claude is offered on several clouds, and features differ between them. Put each agent where its data, identity and compliance boundary already live, then check the features it needs.
Read article →An audit trail for AI agents: when a ledger beats a log
Most agent programmes need a good central log long before they need a blockchain. A ledger earns its cost only when someone outside must trust that a record was never edited.
Read article →Multi-cloud agents without multi-cloud chaos: one control layer
Multi-cloud agent programmes rarely fail on compute. They fail on governance: several logs, several policies and nobody who owns them. Define five things once and apply them everywhere.
Read article →AI agents for government: keeping records where they are allowed to live
Public bodies can use agents without moving regulated records. Keep the data where it is approved to live, bring the agent to it through narrow interfaces, and log every step.
Read article →How to roll out Claude Enterprise: five stages from training to production
Buying licences is the easy part. This is the rollout sequence we use with enterprises — five stages, each with an exit condition — and the decisions that decide whether Claude changes how the work runs.
Read article →What to decide before Claude licences go wide
The governance decisions that are cheap before rollout and expensive after it: approved surfaces, data, logging, approval rights, what an agent may do alone, and who owns each workflow.
Read article →How to roll out Claude Code to an engineering team
Individual engineers adopt Claude Code on their own. Teams do not. A rollout sequence that turns private habits into a shared, reviewable engineering standard.
Read article →How to measure Claude adoption, and what not to measure
Active users and message counts are the easiest numbers to get and the least useful. The measures that show whether Claude changed how work runs — and how to set a baseline before you start.
Read article →Introducing GrabSaaS: find the deal before you subscribe
We built a free catalog of SaaS discounts, coupons, referral offers, free trials and startup credits — every offer recorded with where it came from, when it was last checked, and whether anyone has verified it.
Read article →Stuck in vibe coding? What solo founders need to reach production
The first version took a weekend. The last stretch has taken months, and every fix breaks something else. Why vibe-coded apps stall, and how a solo founder gets one to production without starting over.
Read article →What Claude Code changes in an engineering team
Individual engineers get faster on their own. What changes at the team level is different, and it only happens if you treat the workflow as something to standardise rather than something everyone discovers privately.
Read article →Guardrails: deciding what an agent may do unsupervised
An agent without a defined boundary does not become cautious. It becomes confident. Here is how to draw the line before anything goes live, and what has to be logged either side of it.
Read article →Which workflows to give an agent first
Not every task that could be automated should be. Four properties separate the workflows worth handing to an agent from the ones that will cost you more than they return.
Read article →Why Claude adoption stalls after training
The workshops land, people are enthusiastic, and three months later the work runs exactly as it did before. Here is what actually causes that, and what restarts it.
Read article →Claude Consulting: How Enterprises Choose the Right Partner
A practical guide to Claude consulting — evaluating partners, prioritizing use cases, designing governance, and driving adoption beyond pilot projects.
Read guide →AI Agent Consulting: From Pilot to Production on Claude
What AI agent consulting should deliver — scoping, architecture, guardrails, and production rollout for Claude agents, assistants, and RAG systems.
Read guide →The Azure to AWS Migration Playbook
A step-by-step technical breakdown of migrating a multi-tenant enterprise .NET monolith from Azure App Services to a scalable Amazon EKS architecture.
Read playbook →Zero-Downtime Monolith Migration to AWS EKS
A deep-dive technical breakdown of migrating a multi-tenant enterprise .NET monolith to a containerized Amazon EKS architecture without dropping a single packet.
Read playbook →Building Resilient Agentic Workflows on Claude and MCP
The architecture behind agents that take real actions: Claude for reasoning, an MCP server per system boundary, and a guardrail table written before any code.
Read playbook →Where SMBs should actually start with AI in 2026
Skip the chatbot pilot. Start with the workflows that leak revenue every day — enquiries, follow-ups, and reporting.
Read article →The modern restaurant stack: POS, delivery and AI customer chat
How growing restaurants unify the counter, kitchen, delivery apps, and customer chat on one operating rhythm.
Read article →What digital transformation really means for a growing business
It is not a website refresh. It is replacing manual coordination with systems that scale as you add people, outlets, and channels.
Read article →Automation wins: the 10 workflows worth fixing first
Not every process deserves automation. These ten show up in almost every SMB we audit — and repay the investment quickly.
Read article →Retrieval for Claude at terabyte scale
A million-token context window moved the retrieval problem rather than removing it — routing, re-ranking, parent–child chunking and per-source MCP boundaries.
Read playbook →How early-stage founders choose tech without slowing growth
Buy for the next 18 months, not the next IPO. A practical framework for stack decisions when you are still finding product-market fit.
Read article →Operations as a growth lever — not just a cost center
Tighter operations free capacity for more customers, better margins, and faster expansion — without proportional headcount.
Read article →Securing LLM Endpoints: A Zero-Trust Approach
A zero-trust playbook covering prompt injection, data loss prevention, IAM boundaries and WAF configuration for autonomous agents.
Read playbook →