What I'm watching in 2026

If you can act on it this quarter, it is not a trend. It is news.

That distinction matters more than it sounds. Most trend lists describe things that have already happened to somebody else, which makes them comfortable to read and useless to act on. By the time a shift is legible enough to write a list about, the advantage is gone, and you are just catching up politely.

So I use a narrower test. I watch things where the consequence is already scheduled, and the response is not. Where a date exists in a regulation, a standard, or a supply contract, and almost nobody inside your company has that date in a plan. That gap between a fixed consequence and an unwritten response is the only place a mid-sized company can still get ahead cheaply.

Five of those are running right now. One of them is about AI. That is not an oversight.

Your product data becomes a regulated public record

Start with the one almost nobody in tech is tracking, because it arrives through the product team rather than the CIO.

The EU's Digital Product Passport turns from theory into obligation on 18 February 2027, when batteries go first. Textiles, iron and steel, and furniture have their first delegated acts expected during 2027, which lands them in practice around 2028. Detergents follow in 2029, toys in 2030. If you sell a physical product into Europe, there is a year in that list with your name on it.

Most people hear "passport" and picture a QR code. That is the least interesting part. What it actually requires is that data about a specific item, not a product line, is accurate, current, machine-readable, and available to people outside your company for the life of that item. Composition, origin, repairability, what happens at end of life.

Now go and look at where that data lives in your business today. In my experience, it lives in a product information system built for marketing copy, three spreadsheets owned by people in procurement, and a supplier's email attachment from 2023. It was never built to be authoritative, because until now nothing outside your own website consumed it.

I have been on both sides of this. I built a compliance record system for factory certificates at a company selling procurement software to L'Oreal and Swiss Re, and the hard part was never storing the certificate. It was that the truth was scattered across suppliers who had no incentive to keep it current. More recently, I built a parts finder covering more than 60,000 vehicle types, assembled from scraped and machine-generated data, and I can tell you exactly how confident you should be in product data nobody has ever been held to account for.

What to do in 2026: find out which of your product attributes would survive an external audit and which are folklore. That exercise takes weeks, and it does not get shorter by starting in 2028.

Identity stops being about humans logging in

Every EU member state has to offer at least one digital identity wallet by December 2026. Fewer than a third are currently on track, which is exactly why this is worth watching rather than dismissing. Late regulation does not go away; it arrives badly and suddenly.

The obvious consequence is that verifying who someone is gets cheaper and better, which matters if you do anything with age, credentials, licenses, or anti-money-laundering checks.

The consequence I am actually watching is the other one. At the same time as identity gets solved for people, software agents are starting to act on behalf of people. They browse, compare, negotiate and buy. Ask a simple question about that, and it gets uncomfortable fast: when an agent places an order with you, who exactly signed it? What did the human authorize, how far does that authority go, and what is your evidence if they dispute it later?

Nobody has a clean answer yet. That is precisely why it is on this list. When I moved a retailer to buy-now-pay-later before its competitors and conversion went up 28%, the gain came from being early to a payment shift while everyone else waited for it to be standard practice. Agent authorization is the same shape of question, arriving now.

What to do in 2026: decide whether your systems can tell an agent apart from a human at all, and what you would want to do differently if they could. Not a project. A conversation with your architects that most companies have not had.

Somebody has to write down every place you use encryption

The European Commission's roadmap asks member states to begin the move to post-quantum cryptography with first steps by the end of 2026, high-risk and critical systems secured by 2030, and everything practical done by 2035.

Your instinct on reading that is that 2030 is far away and this is a problem for banks. Half right. The reason it is on my list for this year is the phrase "harvest now, decrypt later". Encrypted data being captured today can sit in storage until decryption becomes possible. So the real question is not when quantum computers arrive; it is whether anything you transmit today still needs to be secret in 2032. For medical data, legal files, long-term contracts, source code and personal records, the answer is yes, and the exposure already started.

The genuinely useful first step is unglamorous, and nobody has done it: a cryptographic inventory. A list of every place your systems use encryption, what algorithm, in which library, owned by whom. Most organizations cannot produce that list, which is a bigger problem than the quantum computer, because you cannot migrate what you cannot enumerate.

What to do in 2026: build that inventory. It is a week of work now and a year of archaeology later. It is also the same discipline as the AI system inventory, and the same people will resent doing both, so do them together.

The seat stops being the unit software is sold in

This one is not regulated, has no date, and will hit your budget harder than the three above.

Business software has been sold per seat for two decades because seats were a fair proxy for value. More people using it meant more work getting done. That proxy breaks the moment the work is done by an agent instead of a person. If a team of 8 does what 20 used to do, a per-seat vendor is now being paid less for a product delivering more, and they will not accept that for long.

So pricing moves toward consumption, or actions, or outcomes. Some vendors have already started. Watch what that does on your side of the table: a cost line that used to be predictable and tied to headcount becomes variable and tied to activity. Your budgeting process, which assumes the first kind, quietly stops working. Finance will notice this before IT does, usually in the form of an invoice nobody forecast.

I spent years building procurement systems for companies like Philip Morris, Nestle and Booking.com, and the pattern is consistent: organizations are good at negotiating price and bad at noticing a change in the unit of pricing. The second one costs far more.

What to do in 2026: check the renewal dates on your five largest software contracts and find out which of them have already changed their pricing model for new customers. That tells you what your renewal will look like before you are in the room.

Who is senior in 2031

Here is the AI one, and it is not about models.

Junior work is the work agents do best. Small, well-specified, heavily reviewed. So companies are hiring fewer juniors, and the ones they do hire spend less time doing the thing that used to make people good.

Now run that forward. Senior engineers are not produced by training courses. They are produced by five to ten years of making mistakes on work that mattered slightly less, watched by someone more experienced. If that apprenticeship largely stops in 2026, the shortage does not show up in 2026. It shows up around 2031, as an absence of people who can tell whether an agent's output is subtly wrong.

I have built teams through this kind of gap before. Growing an organization from 18 to 53 people across three countries taught me that you cannot buy your way out of a seniority shortage, because everyone tries to buy at the same time and the price goes vertical.

I do not have a clean answer to this one, which is honest rather than evasive. What I am fairly sure of is that the companies which keep hiring juniors and deliberately redesign what those juniors do, toward specification, review and judgment rather than typing, will own a scarce asset in five years while everyone else is bidding for it.

What to do in 2026: if you cut junior hiring, be honest that it is a decision with a 2031 invoice attached, and write down who is going to be senior by then.

What I stopped watching

A trend list is only as useful as the things it leaves out, so here are mine.

I stopped watching model releases. For about two years, which model was on top genuinely changed architecture decisions. It stopped changing mine somewhere in 2025. They leapfrog every few months; the differences that matter to a normal company have narrowed, and the correct posture is to make the model a swappable component and stop reading the announcements.

I stopped watching agent framework announcements for the same reason. The winning pattern turned out to be boring: narrow scope, tight guardrails, one workflow owned end to end, a human approving. That has not changed in a year, and no framework fixes an organization that has not decided what it wants.

And I stopped watching benchmark scores entirely, because the only benchmark that predicted anything for me was a fixed set of our own real cases, scored the same way each time.

How to use a list like this

Do not act on all five. That is how strategy documents die.

Pick the one where a date already exists, and your company has no owner for it. For most businesses selling physical products, that is the product passport, and it is closer than it looks. For most software businesses, it is the pricing shift, because it lands on the P&L this financial year.

Then give it a name and a person. Not a program, a person. The gap I described at the start, between a scheduled consequence and an unwritten response, closes the moment someone is accountable for one of these and widens every month nobody is.

I write up what I am seeing every fortnight in Operating Notes, including the parts I get wrong.