Incident sheet · /enterprise-ai

Filed
Read
3 min · 619 words
Severity
Routine

AI coding agents put software on a six-month obsolescence clock

OpenAI's discrete-geometry result, Big Walk's disputed AI-assisted game work and Stefanini's six-month warning point to a shift from coding to validation.

AI agents will soon surpass us
FeedAI agents will soon surpass us — Innovalabs / pixabay

Event log

  1. Five years earlier, Tom Whitfield lost a bet that NFTs would become a major part of gaming by that week.
  2. In May, OpenAI reported that an internal model disproved a central conjecture in discrete geometry: the Erdős unit distance problem.
  3. Fábio Caversan, CTO of Stefanini Group, says code written today will be outdated in six months.
  4. Big Walk used AI to solve a gaming problem a Unity engineer deemed unsolvable; the gaming community apparently considered the approach controversial.
  5. Whitfield expects developers to move from producing code toward QA-style validation as agent output becomes less understandable.

“Whatever you code today will be outdated in six months,” said Fábio Caversan, CTO of Stefanini Group. The warning sets a sharp deadline for enterprise software teams still treating AI as an assistant.

That forecast turns AI coding from an efficiency project into a possible reset of engineering work. My own test came from a family bet: five years earlier, I told my daughter that non-fungible tokens would become a major part of gaming by this week. I lost.

My daughter also told me I knew little about the gaming community, while conceding that my prediction that agentic coding would become the norm was right. Like many young people, she was skeptical of AI generally and disliked the idea of using it to write code. Leveraging large language models to build games drew frowns in the community.

Her view changed after she read about Big Walk, a company that used AI to solve a gaming problem a Unity engineer had deemed unsolvable; the account gives no software version, user count or cost. The gaming community apparently considered the approach controversial.

I initially found that reaction puzzling: the community appeared focused on the harm AI was doing, while I was more interested that AI had solved a problem we had treated as impossible. My conclusion was blunt: everyone must get used to that.

AI has been doing this for some time. Mathematicians have felt it strongly as systems solve problems that stumped humans.

The May result

In May, OpenAI reported that an internal model had disproved a central conjecture in discrete geometry, the Erdős unit distance problem; the account names no model version, deployment scale or cost. The result did not provide a complete answer. It opened a door that humans quickly walked through, reaching a place no one had conceived of.

That sequence matters more than the headline. I expect AI not only to code better than humans, but soon to be orders of magnitude better, including work people have not thought of yet. On that forecast, AI code generation will itself be overtaken by agents that are more functional and produce things nobody has imagined.

The review burden

Today, teams devote much effort to reviewing AI-written code against quality standards. They should prepare for agents whose output may be so different that people cannot understand what they are doing. They may not be far from a world where code stops mattering, not because agents write flawless code, but because their output remains unfamiliar and inconceivable.

What does this change for software teams?

The practical answer is a role change and a harder trust test. If code becomes something humans cannot or need not understand, trust must rest on validation rather than comprehension. The software developer’s job moves close to that of a QA engineer: validating output instead of producing it.

An agent may deliver software that runs and does exactly what we ask, in a manner we no longer care to inspect. The remaining task is proving that the result is correct and what we want.

The Erdős case supplies the operating principle: do not take the agent’s word for everything. Walk through the open door, then check what humans can build from it. For enterprise teams, validation remains central even when generated code becomes less legible.

It also means retiring one assumption. My final verdict is deliberate: “We humans need to realize that when we say something ‘can’t be done,’ what we really mean is ‘no human has done it yet.’”

AI agents are nearing that limit in coding and beyond. The practical issue is not whether they write better code. It is whether organizations can verify results they cannot fully read.

via infoworld.com/feed/ (Source)

Filed under

  • ai-coding-agents
  • agentic-ai
  • software-development
  • code-validation
  • enterprise-trust
Share this article:

More from Tom Whitfield

Tom Whitfield

Show full bio

Market editor covering people, launches and day-to-day practice in enterprise information technology and the leaders who run at CIO Monitor.

183 articles

The newsroom digest

One email a day with the stories that matter. No noise.

By subscribing you agree to the privacy policy.

Cross-references

‹ Previous articleNext article ›