It’s 9pm and you’re working. Again. This time slumped on the sofa with the laptop. Chasing quotes you sent out last week. The kids went to bed an hour ago. You read them half a story before your phone buzzed. If this sound familiar, you’ve a problem with your workflows, not your tools.
Quick recap before we go any further. In Part 1 I said your AI tools aren’t saving you any time because they don’t sit inside a system. In Part 2 we got your data sorted: a clear structure, clear rules on who sees what, and a way to get it back out.
That’s the foundations.
Now we build the walls and the rooms.
Tools sit inside workflows. Not the other way round.
Pick the workflow first. Then pick the tool that fits it. Never the reverse.
This is the part of the series where you actually start getting time back.
It’s also where my AI Agents series and my Cowork Plugins series meet, so if you’ve been reading either of those, this is the bit that ties them together.
Why I Can Say This
I’ve been writing software since the early 2000s and building automation since 2007, from tiny start-ups to big corporates.
I’ve also done it the wrong way round. Loads of times.
I built SaaS products and mobile apps, then went looking for someone with a problem they might solve. AppSumo kicked one of them off for low sales.
Crap, I know. What was I thinking?
I built the tool first and hoped the workflow would turn up later. It never did.
Nothing is ever wasted, though. That failure taught me the lesson this whole post rests on.
#1 – Pick The Workflows, Then The Tools
Don’t Buy The Sofa Before You’ve Got A Living Room
Imagine building a house.
Part 2 was the foundations. Your data. Get that wrong and everything on top of it cracks.
Workflows are the walls and the rooms.
The kitchen is where you cook. The office is where you work. Each room has a job.
Tools are the furniture.
Nobody sane buys a corner sofa and then designs the house around it. You’d end up with a living room shaped like a sofa.
That’s exactly what most businesses do with AI.
They see a demo, buy a subscription, and then try to find somewhere to put it.
Six months later it’s gathering dust in a spare room and the direct debit is still going out every month.
Name The Room First
So before you look at a single tool, write down the workflow.
Not “marketing” or “admin”. Those are whole floors of the house. I mean a single room with a start and an end.
You chase a quote until you win it or lose it. A tender lands in the inbox and you triage it into “bid” or “bin”.
A new client signs and you onboard them. A job finishes and the invoice goes out.
Each of those has a trigger, a set of steps, and a finish line.
Once you can say that out loud, picking a tool gets easy. You’re not asking “what can this AI do?” any more. You’re asking “what does this room need?”
Very different question.
Much better answer.
#2 – Two Kinds Of Workflows
Workflows That Run Themselves
Here’s how I split it. To be clear, this is my model for explaining it, not how any software company describes their products.
Some workflows should run without anyone touching them.
The chasing of a quote is a good example. A quote goes out. Three days later, a follow-up gets sent automatically.
A week later, another.
If they reply, the workflow stops and tells you about it.
Nobody needs to be doing this manually at 9pm to make that happen.
In my model, that’s an agent. It watches for a trigger, does the steps, and only shouts when it needs a human.
Tools like Claude Cowork can do this sort of thing now. You can schedule tasks with /schedule and they run in the cloud, so your laptop doesn’t even need to be on.
There are approval settings too: Manual, Auto, or skip approvals altogether.
My advice? Start on Manual. Let it earn your trust before you take the stabilisers off.
I go deep on these in my AI Agents series.
Workflows That Your Team Runs
Other workflows need a human “in the middle “in the loop”.
Writing a tender response. Pricing a tricky job. Handling a client complaint.
You want judgement there, not autopilot.
These are the workflows your team runs, with the AI sat next.
In my model, that’s a plugin. In Cowork, each plugin bundles skills, connectors and sub-agents into a single package. On Team and Enterprise plans you can share them with colleagues and publish them to your organisation’s library, so everyone uses the same workflows, the same way.
That last bit matters more than you’d think.
Think of a fleet of vans. The agent is the van on a fixed route, doing the same rounds every day.
The plugin is the toolbag in the back that every engineer carries. Same tools, same layout, same standard, whoever’s driving.
I cover building these properly in my Cowork Plugins series.
Why The Distinction Matters
I’ve lost count of the discovery calls where an owner says something like, “We’ve a remote team, it’s great for six weeks, then errors start happening.”
That’s not a people problem. Not really.
That’s a workflow that lives in someone’s head. Everyone does it slightly differently, and it drifts until it breaks.
Some of those workflows should become agents, so nobody has to remember. The rest should become plugins, so everyone does them the same way.
Either way, the drift stops.
#3 – Find The Room Where There’s A Leak.
You Can’t Manage What You Don’t Measure
Right. You know the two kinds workflows.
Which room do you start with?
Not the shiny one. The leaky one.
Every business has a room where the hours drain out through a gap under the door. You don’t notice it day to day. You just notice at 9pm.
So for one week, write down every task you do more than twice. Just a list in a notebook. No app needed.
By Friday, you’ll see it common tasks cropping up.
The same quote chases. Spend hours sorting your inbox out. The same copy and paste between the same two systems.
The Definition Of Insanity
Here’s my opinion, and I’ll stand by it.
Spending hours doing the same thing over and over, to get the same result, is insanity. You’re paid for your creativity, not for your ability to do admin.
And getting your team to do it isn’t much better. That’s like burning money with a lighter, and it’s the reason they end up saying “I don’t want to do that.”
Fair enough. Would you?
Fishing taught me this one.
You don’t cast everywhere. You find where the fish are congregating and you put your effort there. Most of the river is empty water.
Most of your week is too. A handful of workflows hold most of the wasted hours. Find those first.
#4 – Repeatable Beats Clever
The Reps Matter More Than Intelligence
I’m going to offend some people, but repeatable processes beat intelligence. Every time.
A brilliant person doing a task differently every day loses to a “dumb” person whose process is doing it the same way every day. In the gym, the bloke who turns up and puts in the reps beats the gifted one who turns up when he fancies it.
Business and fitness are the same process.
AI doesn’t change that. If anything, it makes it more true.
An agent or a plugin is only as good as the workflow you give it. Give it a clear, repeatable process and it’ll run it all day long. Give it “just sort it out” and you’ll get rubbish back.
Like A Kid At Christmas
I built a report generator for an advisory firm. A workflow they’d done by hand, the same way, over and over.
When I demoed it, the report came out in four minutes.
The couple who ran the business where a husband and wife team. He laughed and said to her, “You’ll have nothing to do now.”
She won’t have nothing to do, obviously. She’ll have the bits that actually need her.
But here’s the thing. That only worked because the workflow was already clear. Same inputs, same structure, same output, every time.
We didn’t make the process clever. We made it repeatable, then handed the reps to the software.
The tool came last. That’s the whole point.
What AI Workflows Won’t Do
I always tell you where an idea breaks. So here goes.
It Won’t Fix A Broken Process
Automate a mess and you get a faster mess. If your quote process is chaos when a human does it, an agent will just make the chaos happen at 3am too.
Fix the room before you furnish it.
It Won’t Work Without Your Data
Everything in this post sits on top of Part 2. If your quotes live in three inboxes and a notebook, no agent on earth can chase them. Foundations first.
It Isn’t Free
Cowork needs a paid plan: Pro, Max, Team or Enterprise. Not a fortune, but not nothing either. Budget for it.
The Split Isn’t Gospel
My “agents run themselves, plugins the team runs” line is a way of thinking, not a law of physics.
Plugins can include sub-agents and hooks. Cowork can run scheduled tasks on its own. The lines blur.
Use the split to decide what a workflow needs. Don’t use it to argue about what a product is.
The Details Will Change
These tools are moving quickly. What’s true about a feature this week might have changed next month. The principle won’t. Workflow first, tool second.
Some Rooms Should Stay Human
Not everything wants automating. The call with a client who’s had a rough week. The handshake on a big job. Keep those. That’s the stuff you got into business for.
Conclusion: Workflows Before Tools
So, to recap:
- #1 – Pick the workflow, then the tool. Name the room before you buy the furniture.
- #2 – Two kinds of workflow. Ones that run themselves (agents) and ones your team runs (plugins).
- #3 – Find the room where the hours leak. Write it down for a week. Start with the leakiest.
- #4 – Repeatable beats clever. Make the process the same every time, then hand the reps to the software.
Your tools don’t give you time back. Your workflows do. The tools just do the lifting once you’ve decided what needs lifting.
Get this bit right and the 9pm kitchen table session starts to disappear. Which means you’re reading the whole bedtime story, not half of it.
In Part 4, we’ll put a roof on the house.
I hope this helps.
Oh, just one more thing.
If you want to see exactly which rooms in your business are leaking hours, take the audit here.
