Execution was never in the model

Every quarter, leadership teams ask the same question: is the strategy landing? Answering it takes days of cross-functional reconstruction. Someone compiles status. Someone else maps it back to strategic intent. Someone else builds the deck for the review. By the time the answer arrives, the question has already shifted.

This past year, we've watched teams try to do that work with AI instead. They feed the activity into a model and ask it to write the monthly report or the QBR. Summarize what shipped, map it to the team's OKRs, format it for the review. The writing is good. Whether the strategy is landing still isn't in the summary.

This isn't a fringe pattern anymore. The 2026 State of AI Agents Report from Anthropic puts report generation as the highest-impact agentic use case beyond coding in enterprises. 65% of technical leaders cite it, 56% plan to scale agents for research and reporting further in 2026.³ What we've been watching in customer rooms is now happening at enterprise scale.

The summary works on what's there. The connections required to answer "is the strategy landing" were never in what the model read. They sit with the people who own delivery, the people who made the commitments, the people who own the dependencies. None of it shows up in the report.

Two years ago, the COO of a major consulting firm told us his organization would not move off strategy PDFs until the next generation of leaders arrived. That position is honest about how slowly large organizations actually move. It doesn't change the structural problem. The connections have to be put there by the people doing the work, AI or no AI.

Last month SAP published the diagram that says the same thing. The CPOs running Signavio and LeanIX call the missing piece Company Memory: "the context, learning, and guidance no enterprise will succeed without."¹ They draw the line at the process layer. The strategic context layer, where choices, objectives, and the teams delivering on them sit connected, is one above. We've been building it for six years.

The next attempt

Some teams take it further. They point the AI at the rest of the stack (Confluence, Jira, Notion, Slack) and ask it to infer the picture from everything around the document. That surfaces what's there. It doesn't produce ground truth. Across 15,000 trials with seven leading LLMs, detailed industrial context shifted strategic recommendations by only about 11%.² The research lines up with what we see.

Inferred connections are the model's guess at what's related, weighted by patterns in the documents it read. The connections we mean work differently: declared by the people who own the work, kept current as the work moves.

Better AI produces better summaries. Whether the connections exist at all is upstream of any model.

What we watch teams do with it

"Are we on track to hit the operating margin commitments we made to the board?" used to be a two-week reconstruction. With the connections maintained, the same question becomes a trace through what teams keep current. The strategic choice connects to the objectives. The objectives connect to the teams. The teams connect through dependencies and commitments. The answer comes back as a traced path: cost-to-serve is down 14% in completed waves, but the self-service platform launch is three weeks behind, and Q4 EBITDA shifts by roughly 60bps if it doesn't resolve this month.

How the answer is traced

What we see unfolding

A team links its objectives to the strategic choices it's serving. Names its dependencies. Marks confidence as it shifts. Every connection one team records makes the picture of how the strategy is landing more complete for every other team. The work is the record.

Better AI will arrive. Bigger context windows. The connections required to answer "is the strategy landing" still have to be put there by the teams doing the work.