By Isaac Correa, Co-founder, Hellomatik
These days, isn’t software the least valuable thing a software company owns? I don’t know how to code, and I’ve still built about 80% of the software I use (an AI-powered Mac app that labels my email, answers it and sets up meetings, a program that manages the terminals where my agents run, and another that pulls options chains straight from Nasdaq and tells me which contracts are worth trading), and I used to pay for a subscription to each of those tools. If someone with no programming background can do that, I’d be surprised if a company with an IT budget doesn’t start asking why it can’t.
Per-seat licensing is where I expect 2027 to hurt first, and Wall Street got there before most IT departments did. At the October 16 close, HubSpot was down 42% year to date and monday.com 39% according to Yahoo Finance, and Forbes (Peter Cohan, back on February 6, 2026) was already calling the whole thing the SaaSpocalypse. HubSpot prices marketing contacts in tiers, so 300 extra contacts can push an account into a bigger monthly bill, and when an agent can do the clicking, the question at renewal stops being how many seats you need and becomes why you’re paying per seat at all.
Most vendors’ answer so far is to bolt an assistant onto a sidebar and call the product agentic. I don’t expect that to hold up next year. Microsoft Outlook and Teams have hundreds of features, and I use a handful of them, so an agent that has to fight the same interface I do isn’t much of an improvement. The agents that matter in 2027 will work directly on a company’s data and on the actions the system already exposes (no buttons in between). That’s also the only setup where you can write down everything an agent is allowed to do and pick which of those it does on its own, then check afterward what it actually did. I already work like that with my own agents: they act, I sit next to them, and every time I correct one, the correction becomes a rule it follows from then on.
All of that falls apart on messy data, which is why I think the real IT project of 2027 is the data layer and not the agent. Ask an agent about a customer who lives in five systems with five different addresses and you’ll get five answers, and nobody in the company will trust any of them. For a large company, a single solution can mean integrating fifteen APIs, setting up its own databases and building pipelines clean enough that the model isn’t answering from bad data. AI writes most of that code now too; what no model hands you is an understanding of a large company’s data, or the trust it takes for that company to open its data to an outsider. So the order I’d bet on is governed data infrastructure across the whole business first, and only then small, specific apps for specific teams on top of it.
Once the system is yours, the data can live on your own machines too, and I’d expect healthcare, defense and transport (which already answer to regulators about where their data sits) to be the first to run agents on infrastructure they control.
Inside companies, the person who builds these tools is going to be one of the most important hires of the year. In a small business that’s one person working with AI, and in a large one it’s someone in every department whose job is turning their colleagues’ pain points into small apps (I call them the company’s Avengers). Hiring doesn’t stop because of that, but people get hired for how they think and how far they can grow, rather than for being fast at a mechanical task a trained system already handles.
Meanwhile the AI app gold rush is heading for a price war. The move I see most often is paying $200 a month for an AI coding tool and using it to build general-purpose software that anyone else can build too. It’s the California gold rush again, where the people who got rich were the ones selling shovels, and when everyone digs with the same shovel and pulls out the same thing, price is all that’s left to fight over. That’s a war I won’t fight, and I expect a lot of look-alike AI startups to lose it in 2027; Peter Thiel was right when he wrote that “competition is for losers.”
Full disclosure: I co-founded Hellomatik in Madrid in 2025, and it builds software for companies’ most complex work, using AI agents and apps that run on each company’s own data, so I have a stake in this prediction. My bet for the end of 2027 is that a good number of companies will have canceled a tool they’ve paid for since the 2010s and replaced it with something their own people built on their own data.
##
ABOUT THE AUTHOR
Isaac Correa is co-founder of Hellomatik, a Madrid-based company that builds software for companies’ most complex work, using AI agents and apps that run on each company’s own data. He writes about the future of software at isaac-correa.github.io, where his essay “The End of Software” is published.
