Is Starting Small With AI Really Worth It?

Hello,

This is Simon with the latest edition of The Weekly. In these updates, I share key AI related stories from this week's news, list upcoming events, and share any longer form articles posted on the website.

P.S. — sharing this issue with a couple of people gets you a free guide. More on that partway down.

If you ask almost anyone how to get started with AI at work, you will hear the same thing: begin with the simple tasks. Draft the email, summarise the meeting notes, tidy up the spreadsheet column. It sounds sensible, it sounds safe, and I have given that advice myself.

But I have started to wonder whether "simple" and "worthwhile" are really the same thing. A task can be easy to hand over to AI and still not be worth the effort of handing it over. So this week I want to look at whether starting small gives you a good return, and what a good return would actually look like.

There are good reasons for the advice. A simple task is easy to explain to an AI tool, so you are likely to get a decent result on the first or second attempt. It is also easy for you to judge the output, because you already know what a good answer looks like. And if it goes wrong, the damage is small, which matters when you are still working out where the tool can be trusted.

There is a confidence benefit too. Someone who has had a few quick wins is much more likely to keep using the tool than someone who tried something ambitious, got a muddled answer and gave up. For a team that is nervous about AI, that early momentum is worth having. It also leaves us with the idea that if we give the simple boring work to AI, we can get on with the more interesting and strategic work that requires our additional knowledge and human insight,

The trouble is that a simple task, by definition, does not take long to do yourself. If a short email takes you five minutes and AI gets it down to three, you have saved two minutes. Once you add the time spent writing the prompt, reading the draft and fixing the bits that do not sound like you, that saving can shrink to nothing, or even go negative.

There is also a hidden cost in checking. AI output needs a human to review it, and that review takes roughly the same effort whether the task was simple or complex. On a big task the checking is a small slice of the time you have saved. On a tiny task it can be most of it.

The other problem is that small wins are hard to see. Two minutes here and three minutes there rarely show up in a budget line or a team's output, so it becomes very difficult to tell your boss, or yourself, that the tool is paying for itself. I suspect this is one reason a lot of AI rollouts feel busy but not obviously valuable.

That said, I would not write simple tasks off. The return changes completely when a simple task is also a frequent one. A two-minute saving that happens twenty times a week is a very different proposition from a two-minute saving that happens once.

Simple tasks also pay off when they are tedious rather than difficult. Reformatting notes, sorting a messy list or turning a rambling thread into a short summary are not hard, but they drain energy and attention. The value there is less about minutes saved and more about the mental effort you get back for the work that needs your judgment.

And there is the learning value. Starting small teaches you how the tool behaves, where it makes mistakes and how to write a useful prompt. That knowledge is what lets you move on to bigger tasks with some confidence, so a modest return on the first task can be a sound investment in the second.

Rather than asking whether a task is simple, I would ask whether it is worth doing. Here are four questions I would put to any candidate for a first AI task:

If you do this for a handful of tasks, a pattern usually appears. The best candidates tend to be repetitive, a little boring and low risk, and they are often a step or two above the very simplest jobs on your list.

To measure the return, keep it modest. Time a task a few times before you start using AI, then time it again afterwards, and note whether you were happy with the quality. A fortnight of rough numbers will tell you more than any vendor's claims.

Starting with simple tasks is still good advice, but only if you treat it as a way to learn rather than as the finish line. The simplest jobs rarely deliver much return on their own, so the goal should be to use them to build skill and trust, and then move quickly towards tasks that are frequent, tedious and still low risk.

If you take one thing from this piece, I would suggest it is to stop asking what AI can do and start asking where your time is actually going.

When analytics on Postgres slows down, most teams add a second database. Then come the pipelines, the sync jobs, and a copy of your data that's always a little behind.

TimescaleDB takes a different approach: extend Postgres instead of splitting away from it. Hypertables partition your data automatically as volume grows. Hypercore compression cuts storage up to 95%. Continuous aggregates keep dashboards live without re-querying everything.

CERN runs Postgres this way for sensor data from the Large Hadron Collider.

No split architecture, no pipeline lag, no new query language to learn. Same SQL, same drivers, same tools.

Start on Tiger Cloud and get $1000 in credits.

Get $1000 Credit

Microsoft has offered voluntary buyouts to about 60,000 employees, around 27% of its workforce, according to Fortune. Eligible staff up to vice-president level have until the end of December to accept, with US severance starting at eight weeks' pay plus continued stock vesting and healthcare. Microsoft's own spokesperson pointed to uptake of 3–5% at other tech firms with similar schemes. The move follows roughly 4,800 job cuts in July and a first buyout round in April, while capital expenditure hit $41 billion in the June quarter.

Why it matters: The pattern is now familiar: shrink payroll while pouring money into AI infrastructure. If you work anywhere near a large vendor, expect leaner account teams and support, and expect your own board to ask why they aren't doing the same.

A Goldman Sachs note reported by Fortune estimates AI could displace up to 9% of workers if productivity rises 15%, with slower hiring pushing the total effect to roughly one in 11. Goldman expects the heaviest losses to come in a recession, when firms accelerate automation. Meanwhile, only 23% of US establishments have built AI into regular workflows, and AI-attributed layoffs have fallen from their spring peak.

Why it matters: The gap between the headline fear and the day-to-day reality is the story. Most organisations are still early, so workforce planning decisions made now, before a downturn forces them, will be the ones that matter.

The Information Commissioner's Office says AI models can themselves contain personal data and that developers cannot justify training on it with broad goals such as "benefitting humanity", according to Computing. Ten developers, including Amazon, Anthropic, Apple, Google, Meta, Microsoft and OpenAI, have made or committed to changes, while xAI is under formal investigation. The ICO has also opened a call for evidence on agentic AI, closing 20 November 2026.

Why it matters: If you deploy third-party AI tools, you are still on the hook for how personal data is handled. The agentic AI consultation is a chance to shape guidance that will land on UK employers.

Thanks for reading, and see you next Thursday.

Simon,

Was this email shared with you? If so subscribe here to get your own edition every Thursday.

You can also follow me on social media: