A cold, grey January afternoon. The meeting room is packed with business stakeholders, enterprise architects and business analysts. The stale air is the sign of hours already spent trying to find the right answer.
The team is looking to establish the way forward for the next generation of the organisation’s integration architecture. The ultimate goal is to streamline the API development so that a business person, who understands the data model, can create APIs without writing a single line of code.
Although it is a well-understood problem space, developing an end-to-end solution is likely to be costly. The effort is estimated at 12-18 months for a development team and costs roughly £2m. The team has also considered adopting an off-the-shelf solution and has found that the cost of customisation is likely to be prohibitive, not to mention the need for a long-term commitment and a hard-to-control cost base.
Somebody asks,
“Can AI help?”
Sceptical looks. A pause. Someone offers that yes, maybe AI could shave two or three months off the timeline. A few heads nod. They move on.
“Can AI help?” might have been the wrong question.
It wasn’t too long ago when the six-fingered, disfigured hands of AI generated human images turned into memes across the Internet. The text generated was gibberish. A few months on, it is, in some cases, impossible to tell if an image is generated by AI.
In November 2025, Anthropic’s Opus 4.5, continuously and autonomously, ran for over 30 hours to develop a complete Slack-like chat application.
Only three months later, Opus 4.6 quietly spun up 16 copies of itself, representing a large development team, with every role from multiple developers, each working on a different part of the project to testers and dedicated agents who focused solely on code quality and the removal of duplication. The team worked uninterrupted for 2 weeks, creating a C compiler in Rust, capable of compiling the Linux kernel.
Read that again. It was one engineer for 30 hours in November, and 16 engineers for two weeks in January, all only three months or one minor version later.
Creating a C compiler from scratch was a challenge comparable to the integration use case mentioned earlier. Coding a full-blown compiler might have taken 12-18 months for a legacy development team. It was completed in two weeks at a total cost of $20,000, which was over 100 times cheaper.
What would be the right questions to ask then?
“Can AI solve this problem?”
We have moved well beyond the days when AI was simply “helping” in the IDE to generate a new function. We can now shift our focus to the entire problem space, instead of a small slice of it. That said, every use case is different, and the answer will not always be a confident “yes”; here comes the real question:
“Can AI solve this problem in 6 months? Or 9?”
Given the exponential pace of progress, delaying such a project by six months or even a year would likely still result in completing it well ahead of schedule at a fraction of the cost.
Does this all sound overly optimistic? We would have thought so too.
But here’s a data point. For one of our clients, we developed a prototype of an end-to-end API design, testing, and deployment solution in 3 days. This would typically take 2-3 months for a traditional dev team to accomplish.
Three days.