Predictions are difficult to make, especially about the future. Or so said Neils Bohr, Mark Twain and me, just now. Making them can be fun though - both in terms of a thought experiment and in terms of looking back at them in a year's time and laughing at how wrong I got it. I was struck by an anecdote in a Bryan Cantrill talk that in 2003 he predicted that Apple would make a phone, it would be the year's must have gadget, they'd call it the iPhone... and that it would be a disaster. This is the mix of propheticness and egg-on-face that I aspire towards.
I am also interested though in looking back at them on a longer scale. As someone who tends to over do self-reflection, it will be interesting to see my headspace versus the industry's and how it evolved. Will I be too pessimistic or optimistic? Will I be a lone voice pointing out a prescient truth or just another member of the peanut gallery (almost certainly)? If I look back at what I thought would happen and what does will I spot things I ignored because it didn't fit my world view?
Ground rules
Every year, until I get bored of it or forget, I will make four predictions and one resolution. The resolution might sound like cheating. To me, though, a resolution is just a prediction you have control over. The predictions will all be based around technology as it is the field I work in and I am passionate about.
I am going to try and stay away from predictions that get too political. I am one of those people - and this will send half of you to the close tab button and the other half to smug agreement, nodding wisely - that everything is political. That said, I am going to avoid politics politics. No red vs blue team, Tory vs Labour, Democrat vs Republican, world revolution vs socialism in one country, nationalised sausages vs a competitve free market for pork produce.
I am going to add an evaluation for each prediction, which will make it easier to see how wrong I am. Largely they will be a bit of fun rather than anything sciencey - "x number of y do z" rather than "a statistically significant of x in the y sector will improve their z by n percent".
The Predictions
1. We will see more companies move away from cloud and back to self-managed
For the last decade or more, cloud computing has been the hot topic. People were tired of maintaining racks of servers themselves. They were fed up of having to hurrily purchase and install machines when their usage grew. I don't blame them! I worked at a video conferencing company at the start of the pandemic and only narrowly avoided being roped in to a day of lifting servers into place.
AWS are very good at what they do, at least for the core services, and have been well rewarded for it. They've made money hand over fist in fact. Azure and GCP also... exist and are used by people. I am sure they are making plenty of money as well. People get a useful service out of it and can avoid a big outlay which can be cruical for smaller, growing companies.
I don't think the momentum can last however. The reason these services are making so much money is because they charge a lot for what they offer. The original promise of the cloud was that yes, it was more expensive than buying a server and powering it, but you saved on the cost of a person or a team in running it. I don't think this is working out. The company you work for no doubt still has these people, they're just called something slightly different and spend time wrangling the nightmarish AWS console rather than physical things.
Obviously this is not an original thought. DHH (creater of Ruby on Rails and CEO of 37 signals) wrote about them moving off the cloud, and the massive saving, this year. I don't think they are going to be an outlier.
I think we will see more of this because of the markets looking more towards profability than growth. Before inflation started to bite we had a long period where interst rates were low, so it was natural for companies to spend big to achieve high growth. Those that did were rewarded handsomely for it. This aligns very well to the cloud: we want to grow as quickly as possible so we need an architecture that can support that. Put another way, we need to spend money on the engineers that can crank out the features needed to bring new customers in rather than on servers which are quickly going to become too puny anyway.
This may well have been a false economy in any case. But with interest rates having risen growth at all costs is not looked upon favourably any more. A one-off, up-front cost is going to be better received than cloud costs which spiral because someone forgot to tear their 50 node test cluster down.
Evaluation
One of the top 250 technology companies in the world by market cap will announce they are moving significant applications off the cloud.
2. Programming copilots will stagnate
I hope I am not overstating things by saying it has been a big year for AI. OpenAI raised $6.6bn, Anthropic got another $4bn from Amazon. Most importantly, Advent of Code was plagued by people gaming the leaderboards with LLMs that solved the problems in less than ten seconds.
Even though Advent of Code is not software engineering that does not mean there aren't people licking their lips at the idea of using it whilst doing engineering or replacing engineers with it. From GitHub Copilot to Claude to Cody it seems everyone is trying to help/replace us engineers!
From personal experience I am pretty lukewarm to these tools. Yes, they're great at boilerplate. Yes, it can help an experienced developer use a new tool indiomatically. But day to day I spend about as much time correcting the copilot as I do generating the boilerplate. Speaking to friends I am not alone.
It is well publicised that AI companies are finding it harder and harder to train models that are a big improvement over the previous one. GitHub Copilot itself is using these models so I expect much the same to happen in the coding space.
GitHub Copilot has recently become free. I doubt this is done from the goodness of their hearts. I do wonder whether the revenue stream has rather dried up and their trying to get more people hooked on it.
Evaluation
This is a bit difficult to judge given there doesn't seem to be much information around about the usage of these tools. Therefore I'll fudge it - I will find articles from the last ninety days on Hacker News that mention Copilot and do a sentiment analysis on them. I will then compare to the last ninety days of 2025. If it's lower I am right.
This year the script got a score of 70.5% positive. Let's see if last year is more negative!
3. More startups will be building with Deterministic Simulation Testing first
DST is something I am pretty excited for. Using testing to find and fix bugs in distributed systems is extremely difficult. So often the edge-cases - particularly the really bad edge case that degrade availability or even lose data - only happen in very specific circumstances. A network partition of a particular set of nodes, a slow disk, a lost packet - any one of these could cause a calamity if it happens at the exact right stage of your algorithm and you didn't forsee it.
DST aims to fight back by making these difficult environmental (disk, network, even time) things deterministic. Rather than using the real thing you simulate them. Time is replaced by a tick in your event loop. A network request or disk write takes some number of ticks. You then introduce pseudorandomness. Maybe 10% of packets get lost. Perhaps every 1 in 1000 fsyncs is pathologically slow. This allows those nasty edge cases to be tested and, because we use pseudorandomness, reproduced every time.
This is extremely valuable, but to do it you kind of need to build for it. Shoe-horning a simulated network and disk into an existing codebase is probably not a go-er. That's why I think it is particularly appealing to start ups at the moment, and I expect that to continue. For example the new SQLite rewrite in Rust, Limbo is built to do it; as was TigerBeetle. For those interested there is a also product - Antithesis - which does the simulation for you at the OS layer.
Evaluation
Five startups will blog about using DST this year.
4. TikTok will be banned in another western country
Skirting perilously close to my no politics rule I predict a western government will ban, in whole or part, TikTok; following the US (whole) and Australia (part: under-16s only). The Australian ban was social media in general but as it's my blog I will give myself some leniency here.
Perhaps I am just at That Age™ now where everything newer than me is wrong but I seriously worry about the effect social media is having on people's attention spans particularly. This is a blog post for another time but this year I have cut down my social media use significantly and my short video clips to zero. It has genuinely made me more relaxed and more able to concentrate.
There is more and more concern amongst law makers, doctors and sociologists about how this might be affecting us. There are of course also the security concerns. It's not a surprise that the two countries where a ban exists, or is planned to, have a not entirely straight forward relationship with China.
Evaluation
Self-explanatory
The Resolution
I will write (at least) one blog post a month
I really enjoy writing blog posts and I am not entirely sure why. There's something about getting your words down that is both relaxing and clarifying. I find this in my day job too; if I am working on a really tricky issue recording my steps and trying to explain them sparks new ideas and realisations.
However, consistency is not my strong point (note to current/potential future employers: unless I am getting paid). It's hard to do something regularly, especially when you are not as good at it as you hope. Trust me, if you think this has been rambling nonsense you should have seen the first draft!
To that end I am going to formalise my schedule (once a month); record it in my physical wall calendar so I have a constant reminder of it and publish a promise to do it (this blog post).
What's next?
Well we just have to wait for all these predictions to come true. About a year should do it. In the meantime I'd love to read a post about your predictions - email them and I'll add them below.