I built a CRM with AI. Then my agents started running the business

A couple of weeks ago, on a wet Saturday lunchtime, I sat at my desk barking orders to Claude while it was quietly building me a sales database in the background. As I was doing this, I was also, on the other side of my desk, losing my sh*t at the printer.

One of us was being systematic and productive. It wasn't me.

That weekend I ended up with a working CRM.

A fully functioning system that consolidated contacts from across my messy lists, cross-referenced data, set up automated updates and connected to other tools, including Mailchimp. I wrote about it, made a YouTube video about it, and asked in the video whether this was agentic AI's ChatGPT moment? Watch it here if you missed it: Claude Dispatch Built Me a CRM. Is This Agentic AI's ChatGPT Moment?

Since then, things have moved on. Blimey. And then some.

From one CRM to a team

Here's what I didn't fully appreciate at the time. Once you have one agent that reliably cracks on with work, the question changes. It stops being "can AI do this?" and becomes "what else could this do?"

Yes, dear reader, I kept going.

Every time a job came back done, I gave it another. And then another. Some of them I gave a name and a job description. I wasn't necessarily setting out to build a team to be fair. I just kept saying yes…

So what I ended up with is a team of agents, each with its own ‘patch’ in business. Finance. Marketing. Sales. Design. Operations. Code. Data. Validation. Security. With a coordinator sitting on top of all of them, routing work and keeping the rest of them on brief. You can’t get the staff.

I then built a brain for the business (insanely powerful in its own right): something they could all call on, and keep updated as they cracked on…‘cos y’know…context.

But the thing that’s surprised me most is how often they now do things I haven't asked for.

A couple of wow moments?

Well, three stand out.

One. Not long after I’d set them up, I got a message from Flo, my finance agent. One of my clients had given me the nod that an outstanding payment was on its way. A few days on, the money still hadn't landed.

Flo had been watching my bank transaction feed, matching expected invoices against what had actually cleared, and noticed the gap. Her message was along the lines of: “X from Y hasn’t paid….but they promised payment on Friday. You might want to chase that."

I’d not asked her to check: she'd done it because that's her job, and flagged the gap before I'd thought to look. Properly surreal.

"This one hasn't been paid. You might want to chase that."

Two. I'd been meaning to get a definitive set of brand guidelines written for years. Voice, tone, colours, type, logo rules, the lot.

Mae, my marketing agent, worked on the tone of voice side by reading back through everything we'd published, every email I'd written, the long-form posts on this very blog, and pulling out the patterns. And Pix, who’d joined the team as design lead, picked up the visual side. Logos, palette, type hierarchy, layout rules, asset inventory.

I was sat outside Tesco when I briefed them, and by the time I’d come home with the shopping I had a single set of brand guidelines. Madness.

Then it got deployed. The CRM, the proposals tool, the client area, the AI intelligence platform. Each surface brought into line with the guidelines in turn. A thing that would, in a previous life, have been a six-month project was done from start to finish between Friday evening and Monday afternoon.

Three. A new business meeting went into my calendar last week. A few hours later, a briefing appeared in my CRM. Company background, the people in the room, the services that might fit, suggested opening questions, things to avoid. No one had asked for it. One of the agents had spotted the meeting, pulled the research together, and filed the brief. INTO THE CRM, PEOPLE…

What it adds up to

None of this is magic. It's just that each agent is doing their bit, reliably, without needing to be reminded (most of the time), so there's less admin in my week, more thinking time, and the client-facing work is getting sharper, not thinner.

Things that were jobs on the long list are now running without me in the background. I'm not going to pretend the team is working perfectly: they need management. But jeepers, the things they’re doing for me…it’s nuts.

Where this goes next

Already clients have started asking how they can do this inside their own organisations. Not the training I do in The AI Advantage, which is about getting leaders confident with AI in their day-to-day. Something deeper. Hands-on and practical. How do you actually get this stuff running for your business?

So that's what I'm working on with a handful of clients now. Two formats. The Agent Advantage, which is the course that picks up where The AI Advantage leaves off. And Agent Action Days, where I come in and we spend the day building the first bits of your agent team together.

This is for people who've done the foundational work with me and are ready to go further.

Interested? Let's have that chat, one-to-one.

Why I’m launching The Agent Advantage

For two and a half years running THE AI ADVANTAGE, I’ve trained thousands of people across hundreds of companies. I’ve had amazing feedback, and saying that doesn’t come easy for someone as reluctant as me.

But I hate hype. And the AI world is absolutely full of it.

Agents have been the topic du jour for the best part of two years. I’ve known for a while that I’d eventually need to build a course giving people the same confidence and practical grounding in agentic AI that THE AI ADVANTAGE has given my clients when it comes to LLMs. But until recently, the hype had created a significant gap between what agents promised and what they could actually deliver.

I’ve believed for a while that agents hadn’t yet had their ‘ChatGPT Moment’: that scary/exciting point when the technology finally begins to deliver on its promise.

Until now. And, like I say, dear reader: I don’t hype things.

Why now?

Over the last couple of days, after some time away doing business planning, I set about trying to build myself a CRM.

I'd been putting it off for months. I'm a one-person operation and I know where my data lives. New prospects tracked in one doc, contacts elsewhere. All very manual, and messy, but functional. The idea of consolidating it all, migrating it somewhere new AND then maintaining a CRM on top of everything else? Not exactly top of the priority list.

So I started a project to figure out how I might sort this.

I've been using Claude Cowork since it arrived earlier this year, dipping in and out, but never quite going all in. This time, I did.

I described my goal: consolidate contact data across sources into a single set, cross-reference them for additional contacts at the same organisations, then pull it all into a spreadsheet.

Cowork cracked on with a fairly complex task. A bit of back and forth later, a consolidated contact list came back. That alone would've taken me a couple of days.

Job done? My curiosity was just getting started.

Why stop there? Could the agent keep scraping those scrapy sources and update the data every couple of days? Done. Cross-reference LinkedIn to make sure I'm connected with everyone on the list? Done. And hang on: I've got agentic coding tools on my computer. Could the same tool that had been working on the data go and instruct another agent to actually build the CRM, rather than me taking out yet another SaaS subscription? Done.

Every task I threw at my agents came back done. Not in a "this'd make a nice demo" way. In a "crikey, what could companies do if they knew how to harness this?" way.

And at that moment, I started plotting the course.

THE AI ADVANTAGE still stands. So many companies are still struggling to get the most out of LLM tools, and there are huge gains left on the table. But THE AGENT ADVANTAGE builds on that foundation and takes things to a different level entirely.

Like I say: I don't hype things and I don't follow trends. My clients trust me, and that's something I take seriously. But I think this is genuinely the most exciting moment in AI since ChatGPT first arrived.

If you want to get this advantage, get in touch. I'm taking bookings now. Let's have that chat, one-to-one.

Dispatch for Claude Cowork: the AI that cracks on while you argue with your printer

There's something properly surreal about sitting at your desk on a Saturday lunchtime, typing orders to an AI that's quietly building you a sales database in the background, while simultaneously losing your mind at a printer.

But that's where we are. And I'm here for it.

Let me explain what's going on.

So, what’s Dispatch, then?

Dispatch is a brand new feature in Claude Cowork, and the headline is simple: it lets you message your computer to crack on with tasks, which it then performs autonomously on your behalf. If you’ve been following the OpenClaw hype, you’ll be familiar with the concept.

Rather than sitting there waiting for an AI to finish a job, you fire off your request from the Claude app on your phone, Cowork gets to work on your desktop machine, and you carry on with your day. When it's done, it's done. A bit like having a capable colleague you can delegate to, without having to hover over their shoulder.

Sounds deceptively simple. The implications are anything but.

My experiment: a sales database, built by barking orders

I've been testing Dispatch by asking it to pull together a comprehensive sales database for me. Not by fiddling with spreadsheets or cutting and pasting from various sources. Just telling it what I want, in plain English, and letting it get on with the graft.

And it's doing exactly that. In the background. Right now, as I type this.

I haven't had to babysit it. I haven't had to worry about whether it's doing what it said it would: it just is.

That's the bit that still catches me off guard, even after a few years deep in this stuff. The shift from "AI as a thing you talk to" to "AI as a thing that gets on with work" is quietly enormous.

Meanwhile, I'm fighting my wife's printer

Now, here's where it gets very human.

Whilst my newly Dispatch-powered Cowork is quietly and competently building my sales database, I have been waging a one-man war against the printer. My wife's. On a Saturday. At lunchtime. When I should probably be doing something else entirely.

I will spare you the details. Let's just say it involves drivers, a cable that may or may not be the right cable, ropey cartridges and at least one moment of genuine despair.

The irony is not lost on me. The AI: calm, systematic, productive. Me: none of those things.

What it does illustrate, I guess, is the point I keep making in my workshops. The goal isn't to replace human effort. It's to point AI at the stuff it's good at, so you can focus your attention where it's actually needed. Or, in my case, where it's needed because the printer won't sort itself out.

What does it mean for businesses?

Think about the tasks in your business that involve gathering, organising, or processing information. The ones that eat hours without requiring much judgement. The ones that sit on someone's to-do list for days because there's always something more pressing.

Dispatch is built for exactly those.

Imagine kicking off a research job first thing in the morning, getting on with client work, and finding the output waiting for you by lunch. No chasing. No watching a progress bar. Just results.

Or a weekly competitive scan. A summary of incoming enquiries. A first draft of a report. All set running in the background as a matter of routine, rather than an occasional scramble.

The businesses that'll get the most from this aren't necessarily the most technically sophisticated. They're the ones that sit down, think carefully about where their time goes, and ask: which of these could I just hand off?

That question alone is worth an afternoon. Preferably one not spent on printer drivers.

Fancy a chat about it?

If you're curious about what Dispatch and Cowork could do for your operations, I'm genuinely happy to dig into it with you. No agenda other than working out whether it's a good fit.

Interested? Let's have that chat, one-to-one.

The Six Hidden Barriers Stopping Your Business From Deploying AI Successfully

According to the ONS (June 2025), just 1 in 5 UK businesses are using AI in their operations as of June 2025. 1 in 5.

Meanwhile, the technology is advancing at breakneck speed. And the companies that have actually cracked AI deployment? They're seeing genuinely transformative productivity gains.

So why the lag? After almost three years working with organisations of all sizes, I've identified six critical barriers holding businesses back.

The good news? They're all addressable.

1. The Sausage Blog Trap

Let me tell you a story that plays out with depressing regularity.

A board feels pressure to 'do' AI. The MD passes the task to marketing. Marketing uses ChatGPT to write a blog about their products (let's say they make sausages). The blog's mediocre.

Conclusion? "AI isn't very good."

Box ticked, AI removed from the agenda.

This is what I call the Sausage Blog conundrum. Companies judge AI based on how well it performs a task they're already brilliant at. When it fails to impress, they dismiss the whole thing. Meanwhile, they're missing where AI could genuinely transform their operations: automating repetitive processes, building custom workflows, or creating tools that solve real operational headaches.

2. The 'Posh Google' vs 'All-Powerful God' Misconception

Ask ten people what AI is, and you'll get wildly polarised answers.

Half will tell you it's just "posh Google". A glorified autocomplete. Nowt special.

The other half think it's an all-powerful digital god, capable of literally anything. They'll ask it to replace their entire finance team. (True story. Someone asked me that…I didn't take that gig on.)

The truth sits somewhere between the two. Without understanding the actual capabilities and limitations of today's tools, organisations are essentially flying blind. And that never ends well.

3. Lack of AI Skills (Especially at the Top)

Right, I would say this , wouldn’t I? But hear me out.

If you don't understand how to craft effective prompts, what features exist under the hood, and what's genuinely possible, you'll never get out of first gear with AI.

Here's what really matters: training has to start with the senior team.

When leadership has proper grounding in what these tools can actually do, organisations become strategically aligned on AI deployment.

Without that top-down understanding, AI remains a departmental experiment that never scales. With it, AI becomes a whole-organisation opportunity.

4. Lack of Imagination

Earlier this year, ChatGPT released major image generation updates. You could create almost any image you could imagine. Within reason.

And what did most people do? They copied the 'AI Action Figure' meme. Take a pre-written prompt, add your name and job title. Done. Zero imagination required. Frustrating, and massively telling.

The tools allow us to fundamentally change how we work. Build entirely new workflows. Transform the actual products and services we deliver. Yet this requires imagination. It requires thinking beyond "what can AI do for me today" to "what becomes possible when I really understand these tools?"

Creatives are phenomenally powerful here. But imagination isn't limited to the creative team. It's a mindset. And to me, it's essential for successful AI deployment.

5. The Curse of Shadow AI

Here's the uncomfortable truth: your staff are already using AI. They're just doing it under their desks, without your knowledge or governance.

And that's a data security nightmare.

Without proper governance and mandated enterprise-grade tools, people will use their own personal accounts. They'll upload proprietary company IP. They might even process customer data through a free AI account. Every bit of that information risks being used as training data for the AI makers.

Here's a key fact: there's a widespread misconception that ChatGPT isn't secure for enterprise use. This is a myth. ChatGPT Business or Enterprise accounts provide full GDPR compliance and data security. Your data is NOT used for training.

But free, unregulated accounts with ANY provider? That's a data disaster waiting to happen.

6. Fear Without Context

Management keeps talking about AI with vague references to "transforming how we work." But there's no training. No explanation of what the organisation is actually trying to achieve. No clarity on what this means for people's roles.

What do they think? "It's coming for our jobs."

The idea of AI replacing skilled workers is vastly overstated in the media. And when that messaging combines with organisational silence, you create fear.

Clarity comes when we talk honestly about AI plus subject experts creating improved ways of working. When we explain that we're freeing people from clunky processes so they can focus on higher-value work that requires their expertise and judgement.

So What Now?

These six barriers are real. But they're not insurmountable. The companies seeing genuine success with AI aren't necessarily more innovative or better resourced: they're just addressing these challenges head-on.

Start there, and you're already ahead of 80% of UK businesses.

…and if you’re in need of a quick summary of all of these points, in musical form….well, here you go:

If you're serious about getting ahead

If you're curious about AI and want practical help making it work for your business, why not talk to me about THE AI ADVANTAGE: a range of workshops and services designed to fit where you're at, whether that's leadership strategy, team training, solving a real-world business challenge, or getting ongoing guidance.

Or all of the above.

Clients have called my sessions "the best workshops I've ever been on" and they're designed to immerse you and your team in how to get the most out of the tools, and how they can fit into your day-to-day operations.

Interested? Let's have that chat, one-to-one.