The AI Directive — Issue #001

The AI tools I actually use every week — and what they’re really worth
Monday, 20th April 2026 | ~6 min read

Welcome to Issue #1. If someone forwarded this to you, subscribe here. If you signed up from YouTube, glad you made it.

THE SIGNAL

The Stack I Actually Use

Someone asks me this at least once a week. Usually over coffee, often slightly quietly, as if they’re expecting me to tell them something that wasn’t in the conference keynote. Which, to be fair, is exactly what I’m going to do.

Here’s what’s in my actual weekly workflow — what I’ve paid for and stopped using, and what I genuinely couldn’t operate without. No ranked list. No affiliate-first selections. Just the stack.

Claude (Anthropic) is my primary thinking partner. Drafting, analysis, working through a decision I need to think out loud about. It hallucinates occasionally and can be over-cautious when I need a direct take — but it’s the tool I’d find hardest to replace. I’m on the Pro tier. Worth it.

OpenClaw is harder to explain briefly, which is why most people haven’t heard of it. It’s an AI agent platform that runs on your Mac and connects your tools — email, calendar, files, APIs. I use it to run automated workflows: monitoring inbound, triggering actions, producing the first draft of this newsletter before I review it. It’s not a chat tool. It’s a workflow orchestration layer. Genuinely different category.

HeyGen handles my video work. I have a digital avatar — looks and sounds like me — that I can deploy for training content and explainers without booking a camera crew. Six months ago it was novelty-tier. Now it’s genuinely usable in professional contexts. I use it weekly.

Zapier is the connective tissue. Every time I think “there must be a way to automate this hand-off,” Zapier is usually the answer. Not exciting. Essential.

What I’ve stopped using: Three different AI meeting summary tools. All of them produced outputs I never read — useful in theory, noise in practice. Also an AI sales prospecting tool that turned out to be excellent at generating emails no one wanted to receive. Both gone.

The honest answer to “which tools should I use?” is: start with one, integrate it properly, and measure whether it saves time or adds complexity. Most AI tools have a gap between their demo and their daily reality. The ones I kept are the ones where reality improved the more I used them.

FIELD NOTES

OpenClaw: What “Agentic AI” Actually Means

Half the AI conversations I have with senior leaders stall on one word: agent. It gets used to mean everything from a chatbot to a fully autonomous system. Here’s a cleaner definition.

An AI agent is a system that can take a sequence of actions to complete a goal — not just answer a single question. Where a chatbot responds, an agent acts.

OpenClaw is the most accessible agent platform I’ve found for practical business use. It runs locally on your machine, connects to your tools, and lets you build workflows that would previously have needed a developer. I use it to automate first drafts of this newsletter, monitor inbound for priority signals, and run research jobs overnight.

It’s not magic — it requires setup and will occasionally do something unexpected. But the productivity ceiling is genuinely high once it’s running.

Verdict: 8/10. Steep initial setup. Reliable once it’s in place.

THE SHORTLIST

1. Anthropic quietly updated Claude’s reasoning capabilities. The difference on complex, multi-step analysis is noticeable. If you haven’t put it on a genuinely hard problem recently, do that — you won’t see the improvement on simple tasks.

2. The UK government published updated AI guidance for public sector procurement. If your organisation sells into public sector, read it. If it doesn’t, read it anyway — it signals where enterprise AI governance is heading commercially.

3. Most “AI strategy” documents I’ve seen are AI aspiration documents. There’s nothing wrong with aspiration — but it’s not a strategy until someone owns it, there’s a budget attached, and there’s a way to know if it’s working. Three tests. Check yours.

ONE THING

The question isn’t “should we be using AI?” That ship has sailed. The question is whether you’re the person in the room who understands how — or the one who has to take someone else’s word for it.

FROM THE EDITOR

If this was useful, forward it to one person who’d get value from it. No referral scheme yet — just the old-fashioned way.

Reply to this email if you have a question you’d like answered in a future issue. I read everything.

See you Tuesday.

— Toby

Keep reading