The demos look incredible. The dashboards look better than they have in years. And somewhere in the back of the room, three executives are wondering why nothing feels stable.
I have been in too many of those rooms. Velocity is up. Output is up. The slides have the curve everyone wants to show the board. And the hallway conversation afterwards is always the same: we are moving faster than we can think, and we cannot tell whether we are getting better or just busier.
Time was never just a constraint. It was an invisible governance system. Slow cycles worked, when they worked, not because anyone designed them to — the delays absorbed uncertainty, surfaced disagreement, and forced decisions to mature before they propagated. AI removes those delays. What breaks first is not execution. It is the structures that were quietly making sure execution was pointed in the right direction.
You Were Never as Slow as You Thought
Ask almost any company that worked well before 2024 why it worked, and you will not get a clean answer. People say culture, leadership, the team. Almost none will say: we had a six-week design cycle, and during week four someone walked into the kitchen and said the thing no one had been willing to say in the meeting.
But that was the system. Time acted as a natural regulator [1]. Scarce development bandwidth created accidental checkpoints where teams had room to course-correct [2]. A two-week sprint produced two weeks of conversations that were not on any agenda; a quarterly cycle, three months of incidental misalignment-detection. Friction did work that appeared on no org chart.
That sounds like nostalgia. It is not. Plenty of those delays were genuine dysfunction — coordination overhead, queueing, handoffs that lost information. Slowness was not good. It was, accidentally, performing a job. Remove it and you do not just gain speed — you inherit the job it was doing.
What Breaks When Execution Compresses
Three things. Not eventually — almost immediately.
The first is alignment. Strategic AI Leader’s 2025 analysis named it Velocity Drift — production accelerating while strategic clarity lags, so the organization ships faster in the wrong direction [3]. In slower environments, misalignment surfaces gradually because progress is gradual. In accelerated ones, divergence travels a long way before anyone notices. The cost of being wrong scales with how fast you can be wrong.
The second is decision quality. Under time pressure, people reduce uncertainty-directed exploration and repeat previous choices [4] — the balance tips away from exploring and toward exploiting what worked before. Add the Harvard EECS finding that AI assistance under time pressure increases overreliance on AI recommendations [5], and the bottleneck migrates from execution speed to requirements clarity and judgment quality [9]. Judgment becomes the binding constraint — and judgment does not scale at the rate throughput does.
The third is the cognitive container. UC Berkeley Haas researchers Ye and Ranganathan, studying a 200-person company over eight months, found generative AI did not free up time — it expanded what workers felt capable of taking on. Prompts at lunch, before meetings, in the evening. Stopping points dissolved [13].
The focus data helps explain what that expansion asks of people. ActivTrak’s three-year dataset found the average focused session down to thirteen minutes, with AI users losing twenty-three minutes of daily focus since adoption [14]. Four compounding stressors — information overload, time pressure, complexity, uncertainty — degrade decision quality together [10], and AI amplifies all four at once.
This is the temporal dimension of Phase Collapse — the distinct cognitive phases of work compressing into one moment. The time buffer was where reflection lived. When it disappears, reflection has to be designed back in, or it does not happen.
Forty-Five Minutes, Seventeen Years
On August 1, 2012, Knight Capital handled roughly ten percent of all U.S. equity trading volume. That morning it deployed new trading software as markets opened — driven by an external date, the SEC’s Retail Liquidity Program launch, which had compressed the testing window. A configuration error meant one of eight production servers was missed. The rogue server began executing trades at machine speed.
Alerts fired within milliseconds. The response chain — identify, escalate, decide, halt — operated at human speed.
Forty-five minutes to stop it. Four million trades. Roughly four hundred and forty million dollars. Market cap dropped seventy-five percent in two days. Seventeen years of profitable operation gone, because oversight was calibrated to a different timescale than execution.
That is the catastrophic version. Most organizations live the small one every quarter and call it being busy.
GitLab is the inverse case. A fully remote company of two thousand-plus employees across sixty-plus countries, GitLab decided early that synchronous coordination did not scale. Instead of adding steering committees, they recorded every decision and its rationale in a public handbook that now exceeds two thousand pages. When AI accelerated the engineering pace, the async infrastructure absorbed it — pacing had already moved out of the calendar and into the workflow. Meeting time runs thirty to forty percent below industry norms.
Between those poles sits the case most readers will recognize. A mid-size B2B SaaS company adopts AI code generation. Velocity triples within three months. Leadership shortens the roadmap by sixty percent and ships fourteen features in the time once allocated for five.
Then the quarterly review lands. Customer adoption is twenty-three percent below projection. Eight of the fourteen features solve problems customers had not prioritized. Velocity looks spectacular; alignment is in a different ZIP code. (Composite case, drawn from multiple reported AI adoption stories.)
The fix is not slowing down. It is shrinking the alignment interval: weekly direction gates, a mandatory two-hour check before each sprint about whether the next batch targets the right problem. Feature-to-adoption ratio rises from forty-three to seventy-eight percent — meaning over half the original acceleration gain had been producing features that never reached adoption. Faster waste, produced more confidently.
IBM’s research lands hard here. Eighty-five percent of AI initiatives fail to deliver expected value, often because data readiness has not been built; when data access takes one to four weeks, decision velocity stalls [15]. The execution layer runs at machine speed; the grounding layer runs at the speed of last quarter’s data warehouse migration. Whatever you save in the first never arrives at the second.
The Quarterly Review Was Always Theater
Sit in any large organization long enough and you notice the rhythm. The steering committee is in three weeks. Everyone runs around. Slides are produced, decks aligned, the meeting happens. The next day, the work continues exactly as it would have without it. The deadline was administrative, not operational. The urgency was performative.
That pattern is not new. AI just made it visible.
When you could only ship every quarter, the quarterly review roughly aligned with the work. Now that real speed is possible — a feature shipped in days, a strategy tested against synthetic stakeholder feedback in an afternoon — the gap between calendar-driven deadlines and operational reality is impossible to miss. Deloitte’s 2026 Human Capital Trends puts it plainly: “the boundary between planning and execution is collapsing,” and leaders “feel overwhelmed — aware of the challenges but struggling to act decisively” [16]. Microsoft Research finds diffusion speed outpaces leaders’ ability to align and reimagine workflows [7].
Corporate cadence is not pure theater. A quarterly board meeting is a synchronization mechanism — twelve independent directors with competing commitments need a forcing function to converge attention. The point is not to eliminate cadence. It is to distinguish rhythms that exist because work needs them from rhythms that exist because the calendar does. The second kind is what AI exposes.
The heading is deliberately forceful. AI makes it easier to see which meetings change the work and which merely keep the calendar moving. One question tells you which is which: would anything material change if this meeting moved by two weeks? If yes, it is doing real work. If no — and you can hear the room exhale when you ask — it is theater.
But Am I Romanticizing Delay?
Three objections are worth taking seriously.
First: I am romanticizing delay. Slowness was never a feature — it was an accident of expensive coordination, and pretending otherwise is sentimentality. Few product managers ever said “let us wait three weeks so assumptions can mature”.
Consider the seatbelt. Early cars were slow enough that crashes were often survivable without restraints. Speed went up, and the safety margin disappeared. The seatbelt is not a defense of slow cars; it is an explicit replacement for a protection slowness used to supply by accident. So is this: the functions friction performed — alignment verification, assumption testing, decision spacing — need explicit replacement now that the friction is going.
Second: AI can replace the safeguards time provided. If one agent generates a plan, a second audits it, a third tests it against historical data, and a fourth simulates consequences, all in seconds, the safeguards are not missing — they run at a different timescale.
That addresses verification: does this output match the spec, are the numbers in range, is the format compliant? Those checks can run at execution speed. Judgment under genuine uncertainty cannot. An AI can tell you that two stakeholders’ requirements contradict. It cannot tell you which stakeholder should win when the conflict is about values. The boundary between AI-checkable and human-judgment safeguards will move. It will not disappear.
Third: if acceleration causes instability, the rational response is managed deceleration. The J-curve seems to support it — EC analysis maps AI adoption with steep productivity declines during a twelve-to-twenty-four-month adjustment [8], and Brookings finds AI investments take two to three years to pay off because complementary assets must be rebuilt [17].
But the evidence cuts the other way. Organizations in the J-curve trough are losing safeguards faster than they can replace them. Orgvue’s 2026 research found thirty-two percent of companies that made redundancies on the promise of AI savings had to rehire — they cut before understanding task interdependencies [18]. That is a sequencing problem with a speed signature, and deceleration does not fix sequencing.
Five Layers You Need Before You Accelerate
If time was the system, this is the system that replaces it.
Direction Lock. Every execution burst begins with a one-sentence intent statement validated against three criteria: it connects to a declared strategic priority, it is measurable within the window, and it would still be the priority if execution took zero time. Re-check at forty-eight hours.
Decision Spacing. Sort decisions into consequential (hard to reverse, cross-functional) and operational (easily reversed, local). Install a two-hour minimum buffer between consequential decisions; cap leadership at three per day. The buffer gives people time to examine a consequential choice before the next choice depends on it.
Cadence Redesign. For every recurring meeting, ask: has it changed a decision in the last three cycles, and would anything break if it moved by two weeks. Meetings that fail both are compliance rituals. The political resistance here is the consulting wedge — the people running the meetings are usually the ones whose status depends on them.
Validation Architecture. Classify validation points as automatable (output matches spec, data in range, format) or judgment-dependent (stakeholder alignment, strategic fit, ethical implication, novel risk). Automate the first at execution speed. Build explicit triggers that pause execution and route the second to a human. Different AI modes — predictive, generative, agentic, embodied — reshape coordination differently, so the boundary moves unevenly across the organization [12].
Transition Sequencing. Calibrate intensity to where the organization actually sits on the adoption curve — and most leaders confuse purchasing tools with integrating them. Early-stage adopters need all four layers at full intensity. Mid-transition can relax decision spacing. Post-transition can pilot reducing direction-lock frequency while keeping cadence redesign and validation architecture as permanent infrastructure.
This is not for everyone. Regulated industries already have most of it mandated; organizations under fifty people coordinate through proximity. And without decision logs, the Pacing Stack is just another compliance ritual.
What I Would Do This Week
Not a transformation program. Three things, each about a day’s work.
Open your calendar. For every recurring meeting in the next two weeks, ask the two questions, then move or kill the theater. The people who get six hours back will thank you; the people who run those meetings will object loudly. Their objection is information.
Pick one initiative and write its one-sentence intent. Pin it somewhere visible. At forty-eight hours, ask whether the work still matches it.
Then look at your headline velocity metric — features shipped, deals closed, deliverables produced — and ask whether it has a directional component. If not, half of what you are celebrating is waste produced with confidence. Activity was never advancement; AI removed the lag that let you confuse them.
Time used to do this work for us, badly but consistently. AI removed the consistency along with the badness. What is left is the chance to do it well and explicitly — or to watch the velocity charts look spectacular until the bill arrives: a misaligned year, discovered three quarters too late.
When time collapses, what breaks first is not execution. It is the structures that ensured execution was moving in the right direction. Direction is the only thing worth executing on.
References
- [1] Wu, C.M., Schulz, E., Pleskac, T.J., Speekenbrink, M. (2022). Time Pressure Changes How People Explore and Respond to Uncertainty. Nature Scientific Reports. https://www.nature.com/articles/s41598-022-07901-1
- [2] Strategic AI Leader (2025). Cognitive Compression in the AI Era. https://www.strategicaileader.com/cognitive-compression-in-the-ai-era-powerful-shift/
- [3] Strategic AI Leader (2025). Cognitive Compression in the AI Era — Velocity Drift. https://www.strategicaileader.com/cognitive-compression-in-the-ai-era-powerful-shift/
- [4] Wu et al. (2022). Nature Scientific Reports. https://www.nature.com/articles/s41598-022-07901-1
- [5] Swaroop et al. (2024). Accuracy-Time Tradeoffs in AI-Assisted Decision Making under Time Pressure. Harvard EECS. https://www.eecs.harvard.edu/~kgajos/papers/2024/swaroop2024accuracy.pdf
- [7] Microsoft Research (2025). New Future of Work Report 2025. https://www.microsoft.com/en-us/research/publication/new-future-of-work-report-2025/
- [8] European Commission Futurium (2025). Seven Feedback Loops: Mapping AI’s Systemic Economic Disruption Risks. https://futurium.ec.europa.eu/en/european-ai-alliance/community-content/seven-feedback-loops-mapping-ais-systemic-economic-disruption-risks
- [9] Strategic AI Leader (2025) and Chalmers et al. (2026). As above.
- [10] Kontogiannis, Kossiavelou (2020). Decision Making Under Stress: Information Overload, Time Pressure, Complexity, and Uncertainty. ResearchGate. https://www.researchgate.net/publication/341708303
- [12] Chalmers, Hunt, Pachidi, Potočnik, Townsend (2026). The Acceleration of AI: Rethinking Organization and Work. Journal of Management Studies. https://onlinelibrary.wiley.com/doi/10.1111/joms.70063
- [13] Ye & Ranganathan / UC Berkeley Haas Newsroom (2025). AI promised to free up workers’ time. Haas researchers found the opposite. https://newsroom.haas.berkeley.edu/ai-promised-to-free-up-workers-time-uc-berkeley-haas-researchers-found-the-opposite/
- [14] HR Executive / ActivTrak Productivity Lab (2026). Focus time hit a three-year low. https://hrexecutive.com/focus-time-hit-a-three-year-low-the-hidden-costs-of-your-workplace-ai-rollout/
- [15] IBM Think (2025). Data delivery delays are slowing decisions more than you think. https://www.ibm.com/think/insights/data-access-delays-slowing-decisions
- [16] Deloitte Insights (2026). 2026 Global Human Capital Trends. https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends.html
- [17] Brookings Institution (2025). The effects of AI on firms and workers. https://www.brookings.edu/articles/the-effects-of-ai-on-firms-and-workers/
- [18] Orgvue (2026). New Research Exposes the Complexity of Deploying AI Systems in the Workforce. https://www.prnewswire.com/news-releases/new-research-exposes-the-complexity-of-deploying-ai-systems-in-the-workforce-302712007.html
Knight Capital case sources: SEC Administrative Proceeding 34-70694; Henrico Dolfing project failure analysis; NYT DealBook (2012). GitLab case sources: GitLab Async Handbook; GitLab IR releases.