Why AI programmes stall in the middle, and what CEOs can do about it
Your AI programme isn't stuck on the tech. It's stuck with managers whose standing rests on today's process. Here's how CEOs change the incentives.
Insights
Field notes from the front line of enterprise AI.
Practical thinking for CIOs and the teams delivering AI in the enterprise, what's working, what stalls, and how to get Claude into production and adopted.
Stop 50 people building the same skill: how to run a company skill library
Shared skills multiply fast, and so do the copies. Run a company skill library: one register, a light review, a named owner per skill and a rule for retiring them.
Which AI work still needs a human sign-off? Set approvals by risk
A sign-off on every AI output builds a queue and a rubber stamp. Set approvals by risk instead: none, one or two people, as Anthropic's Claude Code team does.
Claude Enterprise release review: what to check when features switch on by default
Claude Docs, Slides and Design switched on by default for Enterprise on 15 October. A 30-minute release review for AI Ops leads, plus a free checklist.
How to get an AI connector approved: a security review pack
A connector into an approved system needs a scope review. A plugin with its own backend needs a vendor review too. What to put in each pack, plus a checklist.
AI showback vs chargeback: who should pay for your Claude spend?
Always show back AI spend. Charge back only consumption above a central allowance. The model, the data behind it and when to start.
Claude smart reports: what they show, and where they stop
Claude Enterprise smart reports show what your teams use Claude for, what it costs and where they get stuck. What's in a report, and where it stops.
How to write a steering group update for an AI programme
Your sponsor reads the first paragraph. How to write an AI steering group update that leads with outcomes, owns the stalls and asks for decisions.
What Electric Twin Taught Us About AI, Speed and Expertise
What synthetic research firm Electric Twin taught us: AI lets researchers ask far more questions and act sooner, but it can't supply the expertise.
WebMCP: let agents talk to your website
MCP lets AI models talk to your applications. WebMCP lets them talk to your website. What it is, how we've used it on our AI cost calculator, and the questions to ask of your own site.
How we use Granola at Kowalah
Why I'd give up HubSpot before Granola: how we set it up so our team and our AI agents know what was said in every meeting.
Meet Mabel, how we built our company brain
Company knowledge walks out the door with your people. Here's how we built Mabel, our AI company brain, on Claude Managed Agents, and what we've learned.
Claude Enterprise rollout plan: the first 90 days
A week-by-week plan for the first 90 days after signing Claude Enterprise: what to decide, what to produce, how to drive adoption, and where rollouts stall.
Claude Enterprise Pricing and the Move from Team: What Changes at 150 Seats
Claude Enterprise pricing explained: a seat fee plus usage at API rates, what changes when you pass Claude Team's 150-seat limit, how to model the first-year cost and how to plan the switch (2026).
Claude vs Copilot for Microsoft 365 organisations (2026 guide)
You already run Microsoft 365. Where Claude fits alongside Microsoft Copilot in 2026: Claude inside Copilot, the Microsoft 365 connector, Foundry, UK data terms and the cost overlap.
Choosing a Claude implementation partner in the UK (2026 guide)
A CIO's guide to choosing who implements Claude Enterprise in the UK: Anthropic's partner tiers explained, the four types of provider, and 12 questions to ask.
Claudeforce: the CRM is now the spoke not the hub
Claudeforce puts Salesforce data inside Claude. That means sellers stop logging into the CRM.
How to operationalise an MCP-first product
You can't see your MCP-first product being used. Win back some control, swap tests for evals, support users in their own AI, and plan for agents.
How to design an MCP-first product
An API is a contract. An MCP tool is a contract plus a briefing, read by the model every time it runs. Design your tools around your user's jobs-to-be-done, not your raw API schema.
Why should you build an MCP-first product?
Your users already live inside Claude and ChatGPT. MCP-first meets them there, shifts the cost of inference to your user, and solves the cold-start problem when you launch something new.
10 Ways to take control of your Anthropic spend
Staying under 150 seats to dodge Enterprise pricing keeps you on the plan with the fewest cost controls. Ten levers to govern Claude spend properly.
How to pick the right processes to reinvent with AI
The highest impact decision you can make as you lead your company into an AI future is where to focus. Pick the right domains, linked to your core business strategy and KPIs.
Rebuilding your processes with High Output Management
One of my top book recommendations for CxOs seeking to drive AI transformation across their businesses doesn't mention AI once. The book is High Output Management, a business masterclass written in 1983, by then Intel CEO Andy Grove. The book has a simple (if hard to implement) premise - that as a leader your job is to drive the highest output, at the required quality, for the lowest effective cost.
10 things we're doing to get our website found by AI
The Kowalah website is getting found and cited by AI models such at ChatGPT and Claude. This post shares ten easily copyable tactics you can apply to your own website.
Split Personality: Getting your head around Claude Tag
This week Anthropic launched Claude Tag - an improved agent that your team can chat with in Slack. It was easy to scroll through the release and miss the implications of this change and what it means to your business - in this post I'll help you with that.
How we launched the new Kowalah website in seven days
Today we launched the new Kowalah website with refined positioning, a new style, and a focus on buyer enablement content to help CIOs leading their enterprise AI roll-out. Fifteen years ago I remember being part of a website project that took more than six months. We did this in seven days and here's how.
To build agents - start with a whiteboard and post its
When we start building agents for clients we pick up a pen instead of a keyboard.There is a creative process that comes from having the right people in a room, a whiteboard, and some post it notes.Here are the eight topics we walk through.
Electric Bikes and AI - tackle what wasn't possible before
Your top performers can see AI as lowering the bar on their craft. The truth is the opposite. Like an electric bike on a near-vertical trail, AI turns their own expertise into something more powerful, opening up routes that simply weren't possible before.
How we Architect our Anthropic Managed Agents
At Kowalah, we give our agents human names and personalities. Here’s why we chose specialist agents over one generalist, and why naming them Brian, Krista and Claire outperforms ‘lead qualification agent’ every time.
Anthropic Managed Agents - building the block of flats
The organisations that are having the biggest success with managed agents, don't start by building the agent. They start by building the block of flats that the agents will move into first - the architecture, the plumbing, the electrics...
AI and the future of your people - how to design your org structure for an AI world
In this week's Kowalah Wednesday Webinar, we focused on the big question as AI technology advances: "what does this mean for the people that work in your organisation?" In this article, I'll pull out five images to guide your thinking.
WisprFlow - how I talk to my laptop and stopped typing
WisprFlow has transformed the way that I interact with my laptop and the applications that are on it. I can talk faster than I can type, and AI models love the context that I can provide with long, rambling conversations.
How We Built Karolina: An Autonomous AI Project Manager on Claude Agent SDK and Vercel Sandboxes
Karolina is Kowalah’s autonomous AI project manager, built on the Claude Agent SDK and Vercel Sandboxes. She monitors a portfolio of AI consulting projects, sends daily and weekly reports, answers questions in Slack and email, and updates project data in real time. This post breaks down the four-layer architecture—triggers, orchestrator, agent, and MCP server—and the patterns that made Karolina reliable, fast, and easy to extend.
How to Use Claude Code: Skills, Hooks, and Sub-Agents for Developers
Claude Code isn't just another AI coding assistant. With skills, hooks, and sub-agents, you can build a completely customized development environment. Here's how to go beyond basic prompts.
Enterprise Tech's Reverse Evolution: Files, Fields and Flywheels
Discover how AI agents are reversing enterprise tech trends by prioritizing local file access, unstructured data navigation, and continuous deployment.
Business Process Automation with AI: Turn Process Documentation Into Executable Skills
Learn how Claude Skills turns your business processes into automated workflows that AI executes consistently across your organization.
AI for Leaders: Three Frameworks for Strategic Planning in 2026
Harvard research on 776 professionals proves teams with AI produce breakthrough solutions at 2x the rate. Three practical frameworks for leaders in 2026.
Claude Skills: The Leadership Accelerator You Haven't Discovered Yet
You've mastered AI basics. Now discover Claude Skills - the feature that turns AI from helpful assistant into strategic partner. No coding required, just your process and frameworks.
All I Want for Shipmas: An Enterprise User's Wish List for OpenAI
Last December, OpenAI shipped 12 products in 12 days. Here's what enterprise users actually need from the 2025 edition - and what's probably not coming.
The Secret to Great Prompts: Write Like a Human, Not a Programmer
Most prompt advice focuses on technical tricks. But the best prompts come from clear thinking and good writing - skills you already have.
Gemini 3 Just Launched. Here's What It Means for Your AI Strategy
Gemini 3's benchmarks are impressive, but if your teams are model-hopping after every launch, that's the real problem. Here's how to build depth, not just acces
Build vs Buy AI: When Your Custom AI Platform Becomes an Achilles Heel
A custom AI platform signalled innovation three years ago. Now it competes with vendors shipping every month. How to audit your build and decide what to keep, retire or replace.
ChatGPT Advanced Voice Mode: The Feature Your Team Doesn't Know Exists
Most professionals miss ChatGPT's Advanced Voice Mode. Discover how to use hands-free AI for sales prep, learning on-the-go, and project work—right within your
ChatGPT 5.1 - New personalities put to the test
OpenAI just launched ChatGPT-5.1 Instant and ChatGPT 5.1 Thinking. They come with improved personalities and instruction following. We ask the same question to every personality so you can assess the differences.
Why Longer Business Writing No Longer Works in the AI Era
New research shows proposals doubled in length after ChatGPT arrived—but longer writing now predicts failure, not success. Here's what winning teams do instead.
Prompt Injection Risks in ChatGPT Enterprise: Should You Enable Agent Mode?
Prompt injection is OWASP's top AI security risk. What CIOs and security leaders need to know before enabling ChatGPT Agent Mode and Connectors, and how the same controls apply to Claude Cowork.
How to Share a ChatGPT Project (and What Your Team Can See)
How to share a ChatGPT project with colleagues, step by step: invites, link sharing, permissions, who can see project chats and files, and how Claude Projects compare. Updated September 2026.
Wharton AI Adoption Report 2025: Key Findings for Leaders
Key findings from Wharton's 2025 AI adoption report, the third year of the study: AI use is now mainstream in companies, and three quarters of respondents report positive ROI.
AI‑assisted coding: raising the ceiling and lowering the floor for developer productivity
Engineering teams and their leaders need a clear framework for harnessing AI’s benefits, particularly where experienced engineers have concerns about AI's ability to enhance their developer experience. This guide explains how AI can simultaneously elevate skilled developers by automating tedious tasks and expanding their creative range, and empower non‑technical colleagues to prototype and test ideas.
The State of AI in Business 2025 - MIT report
A recent MIT report suggests that 95% of generative AI projects fail to deliver ROI, and suggests that successful companies focus on buying rather than building their AI solutions.
The AI Leadership Crisis: Even Tech Companies Can't Find AI Executives
Even technology companies are finding it challenging to source experienced AI leadership. In this article, we look at the data and suggest a path forward.
Claude Enterprise vs ChatGPT Enterprise: 2026 guide for UK CIOs
How Claude Enterprise and ChatGPT Enterprise compare in 2026 on security, UK data residency, Microsoft 365, agents and pricing, and how a CIO should choose.
Ready when you are
From reading about it
to running it.
A Vision Map Workshop turns the thinking into a defensible roadmap for your programme, in five working days.