Agents Give You Cadence
Until recently I had not looked carefully at my pension in about eighteen months.
I had meant to. I had several times opened the platform, glanced at the headline number, decided it was fine, and closed the tab. I knew this was not really a pension review. A pension review is a thing where you check the fund allocations, look at the trailing performance against the benchmark, notice if anything is sitting in the wrong place, and decide whether to do anything about it. I had not done that since well before Christmas.
What got me to do it, in the end, was not discipline. It was a fortnightly automated report, written by an agent I have set up for this exact purpose. The agent reads my pension data on a schedule, runs a comparison against benchmarks, writes a short summary, and drops it into my inbox every other Tuesday. The first one was useful. The second one was less useful, because nothing had changed. By the third one, I had a baseline and I could see the small drift. By the fifth one, the report flagged that my European fund was running fourth-quartile against its peer group and had been for a year. I had not seen that, because I had not looked.
I want to make a specific argument in this piece. The conventional sell of AI agents is intelligence — they will think for you, write for you, decide for you. The other conventional sell is speed — they will do in seconds what you would have done in an afternoon. Both are true, in places. Neither, in my experience, is the thing agents have actually changed for me.
The thing agents have given me is cadence.
🌀 Why humans are terrible at cadence
Reliable, repeatable, undistracted rhythm. The fortnightly review you said you would do and did not. The monthly check-in you mean to run on your security posture and never quite get to. The quarterly look at your idea backlog that everyone in your team would benefit from but nobody owns. The annual review of a contract you signed five years ago that has been quietly renewing at a price you would not now agree to.
Humans are terrible at cadence. I do not mean this as a criticism. We are not built for it. The fortnightly pension review competes with a dentist appointment, a delivery that has not arrived, a colleague's draft of a thing they need back today, a customer call that ran long. The dentist wins. The fortnight passes. Next time, the customer call. Next time, the boiler. Six months later, the European fund is still fourth-quartile and you have not noticed.
The standard response to this is to put it on a list. The standard observation about that standard response is that the list does not help, because the reason you did not do the thing was not that you forgot. You did not forget. You were going to do it on Thursday. Thursday happened to be the day the customer's bank-holiday emergency dropped. So Thursday became next Thursday. So next Thursday became "this fortnight," which is to say, never.
The agents do not have a Thursday. They do not have a customer with a bank-holiday emergency. They do not have dentist appointments. They have a schedule, and they execute against it, and they do not feel bad about doing the boring thing because they cannot feel bad about anything. Every fortnight, on the date you told them to, they read the data, run the check, and write the report. The cadence holds.
👥 The reliable colleague
I have started thinking about each of these agents as a colleague, with a specific remit. It helps the design. It also helps when I am explaining them to people who do not work with agents.
The pension agent is my Financial Advisor. He reads my pension data, runs the benchmark comparison, writes the report, drops it in the inbox. He has no other job. He does not also offer me investment products. He does not also try to advise me about my mortgage. He reads, compares, reports. Fortnightly.
My Idea Scout has a different remit. Every two weeks she takes a sweep through the dormant projects in my backlog — the ones that have not had activity in over a month — and asks, for each, whether it is dormant because it should be dead, or dormant because nobody got round to it. The output is a short list with a recommendation. She has saved at least three projects this year that I would otherwise have quietly let die. She has also confirmed a half-dozen that genuinely should die, which is just as useful.
My Cultural Advisor watches for new work from authors I care about — Tom Wright, Elizabeth Bear, Nick Harkaway, a small handful of others — and tells me when a new book lands or a long-form piece appears. I would not check that myself. I am sure I would mean to. I would not.
My Security Analyst runs a monthly integrity audit on the cairn stones — looking for missing fields, broken references, stones that should have been updated and weren't. The kind of unglamorous hygiene work that nobody enjoys and everyone benefits from.
None of these agents is doing anything I could not do. The pension comparison is a couple of API calls and a spreadsheet. The idea sweep is a database query and some judgement. The security audit is a script. The intelligence is not the point. The intelligence is sufficient where it needs to be, and that is the bar.
The point is that the agent shows up. Every time. On schedule. Whether or not I am in a meeting, at the dentist, or on holiday. The fortnightly thing happens fortnightly, which means in the course of a year there are twenty-six reports rather than zero.
📈 Compounding
Twenty-six is a different number from zero in a non-obvious way.
The first report is mildly useful. By the fifth report you have a baseline. By the tenth you can see drift. By the twenty-fifth you can see the second derivative — the rate at which things are changing — which is the thing that actually predicts a problem. That is not a year's worth of insight; that is twenty-six steps along a curve that no human would have walked.
This is where the cadence frame collapses into something larger. There is a piece I wrote a while back called The Long Obedience, a phrase borrowed from Nietzsche by way of fifteen years of building sheepCRM. The argument was that the things that matter compound slowly, in one direction, over years, and that this kind of accretion is what serious work actually looks like.
Cadence is what the long obedience looks like inside a year. Twenty-six reports is a long obedience to the European fund, in miniature. A monthly stone-integrity audit run for two years is a long obedience to a clean data layer. A fortnightly idea sweep is a long obedience to not letting your ideas die through inertia.
The agents do not produce the obedience. I produce the obedience — by deciding what to track, what cadence to track it at, and what to do with the report when it arrives. The agents make the obedience cheap enough to actually keep up. Which, for things humans are terrible at sustaining, is the difference between the practice existing and not existing.
🚫 What cadence is not
I do not want to overclaim. There are several things cadence is not, and three of them are worth saying out loud.
Cadence is not creativity. The fortnightly pension review will not invent a new investment strategy. It will tell you the one you have is drifting. If you want strategy, you still have to sit with the question yourself.
Cadence is not judgement. The Idea Scout will tell me that a project has been dormant for three months. She will not tell me whether the world needs that project to wake up. The judgement is mine. The cadence ensures the judgement gets made.
Cadence is not relationship. I keep wanting to put this in different words because it sounds obvious, but it is not, given how the discourse is going. An agent that reads my pension every fortnight is not a financial adviser in any sense that matters. It does not know my children. It does not know what I am trying to do with my life. It does not know when I will be ready to take more risk, or less. If I ever want a real financial conversation, I want a human across a table. The agent is the thing that ensures the conversation, when I have it, is informed by data I have actually been paying attention to.
The frame I have ended up with is that agents are good at the parts of a job that would otherwise not get done, and worse at the parts of a job that only get done once. Cadence sits comfortably on the first side. Strategy, taste, relationship, and judgement all sit on the second.
🧭 Start with the boring thing
This is connected to The Sum of All Tokens, which argued that the cost of software in 2026 is the sum of token spend plus human insight, and that insight is the scarce half. Cadence is what you get when you spend tokens on the cheap half of the equation — the reliable, schedulable, attention-poor work that humans cannot make themselves do. It frees up the human to spend their attention on the half tokens cannot buy.
It connects, perhaps more importantly, to a thing I find harder to talk about. Most of the discourse about AI agents in 2026 is about replacing work. I am much more interested in the work that is not being done at all — the long tail of "I should really" tasks that quietly erode a business or a life because nobody can sustain them. The pension drift. The dormant ideas. The contracts that quietly renew. The data layer that slowly rots. None of that work is glamorous. All of it compounds. Most of it never happens, because humans are bad at unglamorous compounding work, and we are bad at it for very good reasons — we have lives, and the lives keep happening.
The agents do not have lives. That is the whole point. They show up on Tuesday whether or not it is half-term. They run the comparison whether or not the boiler has broken. They produce the twenty-sixth report whether or not the first twenty-five felt boring. Boring is not a problem for them, which means boring-but-important work finally has a home.
If you are running a small business and trying to work out what to do with AI, I would start there. Not with the thing that excites you most. Start with the thing that should happen regularly and does not. Build the agent that does that thing on a schedule and writes you a short report. Live with it for a quarter. Notice what changes. The intelligence will sort itself out later.
The cadence is the gift.
James Webster is the founder of sheepCRM and director of Croftsware. This piece sits alongside The Sum of All Tokens and The Long Obedience.