Every week another software supplier tells business leaders that its latest system has AI built in, and that now is the time to upgrade.
Some of those upgrades will be worthwhile. The AI inside a system can find an invoice, draft a report or answer questions about a customer's account. But I'm concerned about the assumption behind the pitch: that making use of AI means replacing software your organisation already has.
Think about the employee who moves between the CRM, the finance system, spreadsheets and email. Each system might now have its own assistant. The assistants make individual tasks easier, but the employee still has to join the work together.
An assistant or an agent
That is the difference between an assistant and an agent.
An assistant answers when you ask, inside the one system it lives in. An agent is given a job. It works through the steps, checks its own result, tries another way when something doesn't match, keeps track of where it has got to and comes back when a decision is needed. It can do that across several systems, because it doesn't belong to any one of them.
Take the hour after a sales meeting. Someone has to write up the notes, update the customer record, hold dates in the diary, draft the follow-up email and start the proposal. None of it is difficult. It still takes the evening.
We built an AI colleague for this. It takes a spoken note or the call transcript, updates the record in the CRM, holds the dates and prepares the email, a brief on the client and a first draft of the proposal. After the second meeting it starts from what it recorded after the first. Nothing goes to the client until the person has read it and pressed send.
The person remains in charge without doing the copying.
The kitchen you already have
Picture a kitchen where every appliance has been replaced with a smart one. The fridge knows what's inside it. The oven suggests a temperature. Each is a little better than the one it replaced.
Dinner still doesn't make itself. Someone has to take the food from the fridge to the hob to the oven to the table, and that someone is you.
A cook is a different thing altogether. You say what you'd like for dinner. The cook works in the kitchen you already have, with the appliances you already own, and brings you the plate to taste before it goes out.
Buying each system again with AI inside it is the smart appliance. An AI colleague is the cook. The systems stay where they are. What changes is who carries the work between them.
Before you replace a system
So before replacing an older system, I'd ask whether an AI colleague could already use it. Many systems offer an API, a documented way for other software to read or update records. These connections still need proper permissions, security checks and approval rules, but they may allow an organisation to keep software that already serves it well.
People call this 'headless software'. The software remains the filing cabinet and the rulebook: it stores the information and applies the rules of the business. The screens become optional. People don't have to open them to get something done, because their AI colleagues prepare the work through approved connections and come back when a decision is needed.
There is an exception. Some older systems offer no way in at all, and replacing those may be the right call. The point is to find out before you buy.
That changes what I'd spend money on first. I'd work with the leadership team to identify tasks that cross several systems, test a few with proper safeguards, then train people to direct their own AI colleagues.
That's the idea behind CogniScale.
Each person has a team principal: an AI colleague that knows their role, objectives, projects and team priorities, briefs them each morning and helps with the day's work. Together they build Connected Colleagues for the work that comes round again and again. All of them work through the person's own access to the systems the organisation already has, and prepare work for the person to review. We're developing Workspace Brain to track changes during the day and week and flag what needs a decision.
Most of us spend too much time finding information, copying it into other places and chasing updates. If much of that preparation is already done, there is more room to think, speak to people and make decisions.
Technology starts coming to us with what matters, rather than us going into one system after another to look for it.
The organisations that gain most won't necessarily be those with the newest software. They'll be the ones whose people no longer carry the work from one system to the next.
Three questions for any supplier
So next time a supplier offers an AI upgrade, I'd ask three questions:
- Does your AI work only inside your product, or can it act in our other systems too?
- Can an outside agent read and update records in your system, through a documented connection and under our permissions?
- When your AI finishes a step, who carries the work to the next system: it, or my member of staff?
If you'd like help putting those questions to your own systems, I'd be happy to talk.
Tim Bond is the founder of CogniScale, which helps teams work with AI colleagues and measures what changes.

