AI Potential

What can AI do
for me and my team?

Free up your time so you can focus on your highest value work

AI Potential
01
Title
Why Now  /  The Evidence
AI vs Human Experts

In two years, AI went from 1-in-8 to better than the expert.

How good AI has got
AI win rate against human experts on real professional tasks, May 2024 to April 2026 A rising line chart from 12.5 percent for GPT-4o in May 2024 to 84.9 percent for GPT-5.5 in April 2026, crossing the 50 percent human-parity line in late 2025. 100% 75% 25% 0% Human-expert parity (50%) May ’24 Apr ’25 Aug ’25 Dec ’25 Apr ’26 Win-or-tie rate vs a human expert Frontier model, by release date Above the dashed line, AI beats the average human expert. GPT-4o12.5% o4-mini29.1% o335.2% GPT-539.0% Claude Opus 4.147.6% GPT-5.270.9% GPT-5.584.9%
Source: OpenAI GDPval benchmark (Sep 2025) and subsequent model-launch reports. The line shows the share of 1,320 real tasks across 44 professions where a model’s deliverable is judged at least as good as a human expert’s, blind-graded. 50% means parity with the expert; it measures one finished deliverable, not the whole job.
AI Potential
02
AI vs Experts
Why Now  /  The Shift
Cowork & Codex
The shift

Overnight, building agents stopped being a technical job.

Until recently, creating an agent meant hand-authoring five files. Two new tools changed that, and put the power into plain language.

01 · Claude Cowork
Direct agents over your own work.
Point it at your files, apps and tools, then describe the job in plain English. No code. No five files written by hand.
02 · OpenAI Codex
Agents that build, not just chat.
Hand over a goal and the agent plans, builds and checks the work end to end, the kind of job that used to need a developer.
The opportunity most people miss.
Most teams still point AI at emails and content. The real unlock is putting agents onto the actual work of the business, and letting them run it.
AI Potential
03
Cowork & Codex
Foundation  /  Definition
What Is An Agent
The building block

What is an agent? Five files.

A folder of five files that together make a specialist: a job description an AI follows to do real work. Four things set it apart from a one-off prompt.

01 · Self-improving
It gets better on every run.
Each time it runs it spots ways it could do the job better, and folds them back in. Next month’s agent is sharper than today’s.
02 · Focused
It does one thing, exceptionally well.
A single job, to a high standard, every time.
03 · Autonomous
It runs on its own.
Trigger it with a phrase or put it on a schedule, and it runs without you. Set and forget.
04 · Expert quality
Quality is built in, and you set the bar.
Quality gates live inside its five files and rise over time. Dial the standard up or down to whatever the job needs.
AI Potential
04
What Is An Agent
Part 01  /  Why Agents
Manual vs. Agent
The gap

The gap between doing it yourself
and having an agent do it.

Doing It Manually
  • You do the work from scratch every time.
  • Quality depends on how much time you have that day.
  • Knowledge stays in your head.
  • If you’re busy, the work doesn’t get done.
With an Agent
  • The agent follows a proven process every time.
  • Quality is enforced by built-in gates: it checks itself.
  • Knowledge is captured in five files that persist forever.
  • You say the phrase and the work gets done in minutes.
AI Potential · Agent Builder Sprint
05
Manual vs. Agent
Part 02
The Agent Builder
Part 02

The Agent Builder.

You don’t have to build agents by hand. There’s a meta-agent (the Agent Builder) that takes a description of repeatable work and produces a complete five-file agent package. The tool that builds the tools.

This is the most leveraged thing in the system.
AI Potential · Agent Builder Sprint
06
Part 02 Opens
Part 02  /  Definition
Meta-Agent
Definition

The meta-agent
that builds all
other agents.

The Agent Builder is itself a five-file agent. It lives at 05 Operations / Agents / agent-builder-v6.

You give it a description of repeatable work and a completed example. It produces a ready-to-deploy agent folder. You tell it: “Build an agent for [whatever].” It reads its own five files, asks clarifying questions, then generates all five output files.

What It Needs From YouThree Inputs
Input 01
Description of the work. What does this agent do? What triggers it? Can be a conversation, a doc, or a working example.
Input 02
A completed example. A real instance of the work done well. This becomes the golden example.
Input 03 · Optional
Input sources. Where the data comes from. Files, transcripts, APIs.
AI Potential · Agent Builder Sprint
07
Meta-Agent
Part 03
Agents In Practice

What you actually
do with agents.

An agent is only worth building if it does real work. Here is the fuel they run on, and how a fleet of them fans into a single briefing you can act on.

From scattered data to one decision.
AI Potential
08
Part 03 Opens
Part 03  /  The Fuel
Three Sources
What agents run on

Three sources of intelligence,
structured into one system.

Source 01 · In-house
Internal, but trapped.
It already exists inside the company. It is just scattered across the tools and devices in daily use.
Inside software
CRMProject toolsEmailFinance apps
Inside cloud & storage
DriveDropboxSharePointShared drives
Inside computers
LaptopsDesktopsLocal files
Source 02 · The internet
External, filtered.
Data outside the company. We ingest only the slice that makes us smarter, not the whole internet.
Paid internet
Licensed dataPaywalled researchSubscriptionsMarket reports
Unpaid internet
Public webNewsCompetitor sitesRegulatory
Source 03 · Net-new
Not captured yet.
The conversations nobody is writing down. You have to create this intelligence, not just collect it.
Communication & transcripts
Call transcriptsMeeting recordingsCoaching notesPost-call reviewsNPS feedback
↓
The intelligence OS.
Gobble up all three sources, keep them current, and structure them into a usable form, so the company gets smarter every day without anyone organising anything by hand.
AI Potential
09
Three Sources
Part 03  /  Orchestration
The Synthesis Machine

One orchestrator. Ten chains. One briefing.

How it scales
One orchestrator spawns ten function chains that fan into one synthesis machine An orchestrator spawns ten parallel function machines, each a collect-extract-scribe-brief chain. All ten briefs fan into a synthesis machine that produces one executive brief for a human. FUNCTION COLLECT EXTRACT SCRIBE BRIEF Orchestrator spawns 10 CollectExtractScribe CollectExtractScribe CollectExtractScribe CollectExtractScribe CollectExtractScribe CollectExtractScribe CollectExtractScribe CollectExtractScribe CollectExtractScribe CollectExtractScribe BriefBriefBriefBriefBriefBriefBriefBriefBriefBrief StrategyProductMarketingSalesFinanceOperationsIT / DataHR / AdminLegalImprovement Synthesis machine GATHER · RANK · COMBINE One executive brief what changed · decide ● Human · decide
AI Potential
10
Synthesis Machine
01  /  10
Pre-flight check

Slide-by-slide report.

– · – × –
Running…