AI works best as a team sport.
AI has been adopted widely and mostly individually. Each person is using it to speed up their own work. 75% of knowledge workers now use AI.1 But 78% of them are bringing their own tools, outside any shared system their company provides.
And only 4% of companies report meaningful ROI from AI adoption.3
The adoption is there. The coordination is not.
The essence of AI is collective. Shared context. Coordinated decisions. Alignment across a team. That is where the value actually is and that is what this article is about.
In today's newsletter:
What Heathrow Terminal 5 was, and why it is famous as a project
The three elements behind the Single Model Environment
What happened when 776 P&G professionals used AI as a team
What I built by hand and where AI can be useful
What Heathrow Terminal 5 Actually Was
Courtesy RSHP
Heathrow Terminal 5 (T5) is the fifth terminal at London Heathrow. Construction ran from 2002 to 2008 at a cost of £4.3 billion. It is one of the most studied mega-projects in construction history not because of its size, but because of its outcome.
Every major UK construction project over £1 billion, and every international airport opened in the prior 15 years, had failed to deliver on time, on budget, and at quality.
T5 delivered all three.
Most people assume the reason was superior engineering, better tools, or smarter contractors. It wasn't. The engineering was competent, not revolutionary. The contractors were the same firms that had failed on other megaprojects.
What changed was how 50,000 people on the project site shared information, through a system called the Single Model Environment.
The Single Model Environment, Three Elements
The Single Model Environment (SME) was not software. It was not a file. It was a methodology, a set of rules for how information moved between people. One continuously updated representation of the terminal, accessible to everyone building it.
No "my version" and "your version." One version.
Three elements made it work.
One shared source of truth: One continuously updated representation, no separate versions
Mandatory participation: Written into every contract, non-negotiable
Co-location and real-time access: client, contractor, and design team on site together, working from the same model in real time
The result: a 5% cost reduction against target.
But the real result was coordination. 50,000 people could make decisions without re-litigating whose version of the truth was correct.
Most organizations cannot currently do this with AI. The coordination layer is missing and that is where the value is hiding.
What Happened When P&G Used AI as a Team
In 2024, Harvard Business School, Wharton, and Procter & Gamble ran an experiment with 776 professionals. They were assigned to develop new product ideas, some individually with AI, some in teams with AI, some without AI at all.
The individual finding was predictable. People working alone with AI were 16% faster. That is the productivity gain everyone talks about.
The collective finding was different.
To reach the top 10% of performance, the winning combination was a full human team plus AI. Not individuals plus AI. Teams plus AI.
And AI did something specific in those teams that no individual tool could do, it broke down silos between R&D and commercial staff. Junior people with AI contributed at near-expert levels. The typical dominance of louder or more senior voices was dampened.
AI acted as a leveler inside the team. Not just an accelerator for individuals.
The Alignment Doc That Worked Until It Didn't
I built a version of this for my own design team. Without AI. By hand.
The design team's role was blurred from the C-suite — leadership could not clearly see how design connected to business outcomes. We were at risk of losing people not because the work wasn't valuable, but because the value wasn't visible.
The problem was that nobody could see how the design team's work connected to company objectives.
So I built a simple alignment doc the designer's work and their impact were visible
Each row linked directly back to the corporate annual objectives. The C-suite could now see — on one page — what every designer was doing, why it mattered, and how it connected to the numbers they already cared about.
It worked. The doc didn't just protect headcount. It became a tool for deciding what to stop doing. Once every project had to justify itself against the company's own objectives, low-impact work became impossible to hide.
But it didn't scale.
When projects shifted or people changed teams, the doc went stale unless someone actively updated it. The methodology was right. The maintenance was entirely manual.
This is an example of where AI should be — not speeding up one person's work, but keeping a team aligned.
Pulling project context from across people, keeping the shared picture current, surfacing where work maps to outcomes and where it doesn't. That is AI for team alignment, not individual productivity.
This tool doesn't exist as a product yet. But it can be built with AI agents today — the building blocks are there. What's missing is someone packaging it for teams, not engineers
Key Takeaway
AI adoption is widespread. AI coordination is not.
T5 worked because of shared information, not superior engineering.. The first answer comes faster and looks more polished, so we question it less.
The Single Model Environment rested on three elements: one shared source of truth, mandatory participation, and co-location with real-time access.
Sometimes minimum alignment documentation is enough. We don't always need a system as sophisticated as SME.
AI Tools I'm Using This Week
💬 Claude — for turning scattered team notes into living shared memory across projects.
🎥 Loom — I use this to explain processes with video recordings instead of writing docs. Faster to record, easier to follow, and the context doesn't get lost in translation.
Note: Questions on AI workflows and implementation in your product or design team? Reply to this email.
That's all for this week. See you in the next one.
1 75% of knowledge workers use AI — Microsoft 2025 Work Trend Index
2 78% use their own tools, not employer-provided — same report, same link above
3 4% of companies see meaningful AI ROI — this one came from your original research notes attributed to Atlassian, but I don't have a direct URL for it. Worth verifying before publishing.