A couple of years ago, a typical two-day task for an analyst, associate, or product manager looked something like this: one and a half days spent collecting data, running desk research, pulling together the right documents, cleaning spreadsheets, and building a first-pass analysis. Then half a day — maybe less — synthesising findings and forming an early recommendation. A three-to-one ratio, heavily weighted toward the doing.
Nobody designed it that way. It was simply the physics of the job. Research takes time. Access to information is imperfect. You have to read before you can think.
In 2026, that physics has changed. The ratio has flipped.
Section 01 — The Old Equilibrium
To understand what has shifted, it helps to be precise about what the old model actually looked like in practice — not in theory, but on the ground.
| Phase | Activity | Time Allocation | Value Delivered |
|---|---|---|---|
| Phase 1 | Data collection, desk research, document review, cleaning | ~1.5 days | Foundational — necessary but not differentiating |
| Phase 2 | Synthesis, pattern recognition, recommendation formation | ~0.5 days | High-value — this is what managers actually needed |
The uncomfortable truth embedded in this table: 75% of the time was spent on work that, while necessary, was not the work that managers actually valued. The recommendation — the "so what" — was being produced in the last hour of a two-day effort.
Nobody talked about this explicitly. It was simply the cost of information asymmetry in a pre-AI world.
Section 02 — What AI Actually Changed
AI did not eliminate the need for research. It compressed it. Dramatically.
A competent analyst using Claude, Perplexity, or a well-configured research agent can now complete in two to three hours what previously took a day and a half. The research is not worse — in many cases it is broader and more systematically organised than what a single human could produce under time pressure.
"Managers now expect the 'doing' part to be over in half a day. What they want is the other one and a half days spent on what they used to do."
That last sentence carries more weight than it appears to. The expectation is not simply that associates work faster. The expectation is that associates move up the value chain — into pressure-testing assumptions, checking for blind spots, expanding the solution space, and building a more thought-through recommendation through iterations.
This is work that previously sat with senior managers and partners. The bar has moved.
Section 03 — The New Distribution
| Phase | Activity | Time Allocation | Delta |
|---|---|---|---|
| Phase 1 | Data collection, research (AI-assisted), initial structuring | ~0.5 days | ↓ 66% |
| Phase 2 | Critical review, assumption pressure-testing, synthesis, iteration | ~1.5 days | ↑ 3× |
The total time has not necessarily shrunk. What has changed is where the time goes. Phase 2 — the thinking, the questioning, the iteration — now consumes three times as many hours as it did before.
This is not a comfortable shift for most associates. Research has clear completion criteria. You can measure it. You know when you are done. Thinking is harder to bound. It requires tolerance for ambiguity, a willingness to tear up a recommendation and rebuild it, and the confidence to push back on your own conclusions.
Associates who treat AI as a speed-up tool for Phase 1 — without expanding their Phase 2 effort — are optimising for the wrong variable. They deliver faster, but not better. The manager notices, even if they cannot immediately articulate why the recommendation feels thin.
Section 04 — The Communication Gap
Here is the problem that nobody is naming out loud: most managers have not told their teams that the bar has moved.
The expectation shift is implicit. It lives in the slightly longer pause before a manager says "good work." It shows up in feedback that sounds like "you need to go deeper" or "I expected more pushback on the assumptions" — without any explanation of what "deeper" now means given that research no longer takes two days.
Associates, operating without this context, continue to optimise for what they can measure. Thoroughness of research. Coverage of sources. Formatting of the deck. The observable outputs of Phase 1.
This dynamic is not anyone's fault individually. It is a structural communication failure — a lag between the change in the operating environment and the update to the social contract of the job.
Section 05 — What To Do About It
The fix is simple in principle and requires deliberate effort in practice. It operates at two levels.
If you are a manager
You have probably already updated your internal model of what "good work" looks like. The question is whether you have made that model explicit to your team. Here is what that conversation needs to include:
- 1.State the new ratio directly. "AI has compressed research time. I expect you to use the saved hours on synthesis, not on producing a more exhaustive literature review."
- 2.Define what Phase 2 looks like in your context. For a PM, it might be identifying the three assumptions most likely to invalidate the roadmap. For a consultant, it is locating the 20% of the recommendation that your client will push back on hardest.
- 3.Create space for iteration. If your review process does not allow for "I tore up version one because the core assumption was wrong," you are implicitly rewarding Phase 1 optimisation.
If you are an associate
Ask yourself honestly: when AI finishes the research in three hours, what do you do with the remaining five? If the answer is "polish the slides" or "find more sources to be safe," the ratio has not actually flipped for you — only the speed has changed.
Take your last completed deliverable. Identify the three core assumptions it rested on. Now pressure-test each one: what would have to be true for this assumption to be wrong? What data would prove or disprove it? If you cannot answer this in under ten minutes, Phase 2 was underinvested.
- Spend the first 30 minutes of any project defining the problem and its key assumptions — before starting research.
- When AI returns a synthesis, treat it as a first draft to interrogate, not a final answer to format.
- Build in at least one deliberate "tear-up" — a moment where you ask: if I were starting from scratch with what I know now, what would I recommend?
- Make your thinking visible. Show the manager the assumptions you challenged, not just the conclusions you reached.
Section 06 — The Larger Shift
The ratio flipping is not an isolated event. It is one instance of a broader pattern: AI is compressing execution and expanding the surface area of judgment. The work that remains — and that will remain — is the work that requires you to know when the model is wrong, when the framing needs to change, and when the right answer is to question the question.
This is not easy work. It is, however, the work that will separate strong performers from average ones in the next five years. The researchers who optimise for research will be replaced. The thinkers who use research as raw material — and who know how to direct AI toward the right problems — will not.
If you are a manager — have you explicitly told your team what the new bar looks like?
If you are an associate — are you spending your hours on deeper thinking, or on faster doing?