Why every software platform will have to let customers build on it

Last week a friend who builds technology for financial markets asked me whether the big cloud providers will lose share now that agents make infrastructure so easy to build. We ended up on a bigger question: what happens to every software vendor once its customers can build software as fast as the vendor can?

I've never seen technology move this fast. ChatGPT came out at the end of 2022, and we've already gone from a chat window to agents that write and ship production code. This is the most basic AI will ever be again.

Customers can build the tool themselves

Take the Bloomberg terminal. It's one of the most successful software products ever sold, and its customers are quants and fund managers with very specific ideas about the tools they want. Today those ideas become feature requests that wait for a roadmap.

Picture the alternative. A quant describes the analysis they want, an agent builds it, and three minutes later it shows up on their terminal next to everything else they use. The fund owner who wants a custom risk view has it the same afternoon. Nobody files a ticket.

Hardware is moving the same way. Companies are now shipping programmable machines designed for agents. A fund could set up a fast box, a good keyboard and two monitors, describe what it wants, and have its own terminal software running in a short time. What it couldn't build is the market data feed. The software is the part that's become easy.

That leaves the vendor with a choice. It can let customers build their own capability inside the product, or it can watch them build it somewhere else.

Work management software faces the same pressure

MetaSpark sits in a category where this pressure is obvious. For a decade, work management tools have asked teams to fit their work into the vendor's boards, fields and automations. When a team needed something the vendor hadn't built, it exported to a spreadsheet and did the work by hand.

We built MetaSpark on the assumption that every team's work is specific and the software should adapt to it. Agents do the work across the systems a team already uses. They pull status, cross-check records, draft the update and rank what matters today. When a team needs to reach a system we've never seen, the connector compiler writes the integration during the conversation. The new capability shows up where the team already works, the same way the quant's tool shows up on the terminal.

Business teams and IT build together

Speed raises a fair question, and IT is usually the first to ask it. CIOs want an agentic strategy as much as anyone in the business, and they've built their organizations on the governance and compliance that make one safe. Before an agent built by a business team goes anywhere near production data, both sides need clear answers. Where does the agent run? What can it reach? Whose permissions does it carry? Who approves it?

A platform that lets customers build has to answer those questions inside the product. In MetaSpark, agents work within the same board and workspace permissions as the people they work for, and connectors reach only the systems an administrator has connected. That lets business teams and IT work together: the team moves quickly, and IT keeps the controls it needs.

What this means for people starting their careers

This has a harder side. Agents already handle a large share of entry-level knowledge work out of the box. A graduate with a master's degree now competes with an agent for the first rung of the ladder. Experts are still in demand, and their judgment is worth more than ever, but the relay work that used to train junior people is moving to software.

I think the answer is to teach people to direct and build with agents from their first day, and to design the work so their judgment gets captured and reused. Companies that do that will grow their people faster, and the software they run on should make it possible.