Five minutes. Two measurements, one picture, four functions built, one board.
“I want to run the business, not do all the work in it.”
“I do not want to become the person who fixes the technology.”
“Whatever you show me today is out of date in six months.”
Run the business, not the work in it. Never the person fixing the technology. And a system that keeps up on its own. Everything that follows is measured against those three.
Source: OpenAI, GDPval report of 25 September 2025 and paper arXiv 2510.04374 (Figure 5: 220 tasks, blind pairwise grading by professional industry experts, wins plus ties); OpenAI, GPT-5.2 page, 11 December 2025 (70.9%, “according to expert human judges”); OpenAI, GPT-5.5 page, 23 April 2026 (84.9%; GPT-5.4 83.0%, Claude Opus 4.7 80.3%; judge not stated; drawn hollow). All figures are OpenAI-reported. Points sit at release dates. OpenAI’s own leaderboard is no longer active. The next slide uses a second, independent measurement.
METR measures how long a task can be, in expert time, and still be completed by AI half the time. The tasks are mainly software work.
GPT-4, March 2023. A question with an answer.
Claude Opus 4.6, February 2026. A piece of work you hand over and come back to.
How often that task length has doubled since 2023.
Source: METR, Task-Completion Time Horizons of Frontier AI Models, metr.org/time-horizons, data file v1.1. 50% success horizon: GPT-4 3.99 minutes; Claude Opus 4.6 718.8 minutes (11.98 hours), METR’s confidence range 5 to 61 hours; doubling time since 2023 128.7 days. Latest published figures: the page was last updated 8 September 2026 and says it is no longer actively updated. Measurements above 16 hours are outside the range METR calls reliable. A time horizon is task difficulty in human time, not how long the AI runs.
Hand it the whole task: the goal, the material, the standard. It drafts, analyses and builds the finished piece.
Your insight into the business, your expertise in the job, and your judgement on what good looks like.
Work that fits your business, and your time back for the thinking only you can do.
51 projects from 47 expert frameworks. Find the reason that sounds like your company, open it, and copy the project.
52 projects from 44 expert frameworks. Find the reason that sounds like your company, open it, and copy the project.
54 projects from 44 expert frameworks. Find the reason that sounds like your company, open it, and copy the project.
53 projects from 44 expert frameworks. Find the reason that sounds like your company, open it, and copy the project.
Every piece of work in one place: what is live, what is waiting on you, what is waiting on someone else, and what the agents are doing right now.
It arrives the first time you open your AI assistant each day. Your meetings, your inbox and your notes have already been read and turned into cards overnight.
Each card is a yes, a no, or a redirect. Anything sent outside the business in your name waits for your click. Nothing is sent on its own.
This is ours, as of this morning.
We help businesses use AI safely: the brain, the rules and the board, built around your numbers, with your people doing the work and your judgement deciding what ships. When we leave, the capability stays.
Install AI capability inside your business: the numbers that matter, the projects that move them, and the agents that do the work.
People first. We show rather than tell. Nothing leaves your business in your name without your click, and we never touch your data.
Book a conversation at aipotential.ai. Bring one number that is stuck; we will show you what the wall looks like for it.
AI Potential · aipotential.ai · 01273 011482