AI Can Build Software in an Hour Now: What It Means for Your Business

In June 2026, an AI coding agent reportedly built a full game clone, multiple environments, day-night cycle, core mechanics, in about an hour, from a single prompt. The same month, Microsoft researchers showed coding agents cutting their own token use by roughly 60% on software tasks while scoring better, not worse. Take the two together and a trend snaps into focus: the cost and time of turning an idea into working software is collapsing. For businesses, that is not a developer story. It is a strategy question, because when building software gets cheap, the math on what you build, buy, and automate changes.

What actually changed

Two things moved at once. Capability: AI coding agents can now one-shot non-trivial, self-contained projects that would have taken a developer days. Economics: the same work is getting cheaper to run, as efficiency research like Microsoft's FastContext cuts the tokens (and therefore the cost) a coding agent burns. When capability rises and cost falls simultaneously, you get an inflection, not an incremental improvement. The friction that used to make custom software expensive, time and specialized labour, is dropping fast.

The important nuance is what has not changed. A one-hour app is a brilliant first draft, not production software. Security, reliability, integration with your real systems, and long-term maintenance still require human judgment. What collapsed is the cost of the first 80%, not the discipline needed for the last 20%.

This is a productivity story, not a replacement story

It is tempting to read "AI builds software" as "AI replaces developers." The reality on the ground is different: technical people become far more productive, and the bottleneck moves up the stack, from writing code to deciding what is worth building, integrating it safely, and supporting it. The teams getting the most from this are pairing AI's speed with human judgment, not betting on one to the exclusion of the other, the same pattern we described in AI agents leaving the demo stage, where the value is in the system and oversight around the model.

What it means if you are not a software company

Most of our clients do not sell software, they run businesses that use it. For them, the cheap-software shift shows up in two concrete ways.

1. Custom internal tools finally make sense. Every business has workflows that no packaged product fits, the spreadsheet everyone hates, the manual step between two systems, the report someone rebuilds every week. Those were never worth a custom build at old prices. Now they often are. The opportunity is to automate the specific, unglamorous processes that off-the-shelf software never quite handled, exactly the work we map in our AI automation playbook and the automation finder.

2. The bar for bought software rises. When capable tools can be assembled quickly, generic products lose some pricing power and you should expect more tailored, cheaper options, and feel freer to walk away from software that no longer earns its cost. This pairs with the build-vs-buy thinking in our piece on buying vs building customer-service AI agents.

The discipline that still matters

Cheap creation makes a new failure mode easy: shipping AI-generated software that looks right and quietly is not. AI can produce code that mishandles edge cases, leaks data, or breaks under real load. Anything customer-facing or touching sensitive information needs human review, testing, and, in Canada, attention to your PIPEDA obligations. Treat AI-built software as a strong draft from a very fast junior developer, invaluable for speed, never shipped unreviewed.

How to take advantage in 2026

The winning move is to make building cheap enough to experiment, while staying disciplined about what you put into production. List the repetitive workflows no tool handles well; prototype the most promising quickly to prove value; then invest in hardening only the few that earn their keep. You get the upside of collapsing software costs, more tailored tools, less manual work, without inheriting a pile of fragile, unmaintained apps.

The bottom line

When an AI can build a working app in an hour and keeps getting cheaper at it, the constraint on software stops being "can we afford to build this?" and becomes "what is worth building, and can we run it responsibly?" That is a far better problem to have. Businesses that lean in, automating the workflows packaged software never fit, while keeping human discipline on security and maintenance, will turn cheap software into a durable operational edge.

Frequently Asked Questions