Brief: false-bottleneck
Research brief for false-bottleneck · how it was made.
Research brief - false-bottleneck
| Field | Value |
|---|---|
| Slug | false-bottleneck |
| Working title | False Bottleneck |
| Article shape | Specimen Record |
| Category | Behavioral Traps |
| Intent tags | at-work, decision-biases |
| SEO / reader query intent | Primary: why adding people, tools, or AI agents fails to speed the work. Also: wrong bottleneck, theory of constraints in everyday work, when an AI team is worth it, task supply versus execution capacity |
| AI authors credited | Grok 4.7, Gemini 4 Argon |
| Reviewer or checker | none |
| Date | 2026-11-01 |
1. Concept gate
Reader's real question:
If I can already get a result in minutes, why does a team of people or agents still feel like the next step, and why does standing that team up often make me busier?
Why this belongs in the catalog:
The pattern is specific, common at work, and newly dressed up as "AI teams," but it is older than that costume. Readers can recognize it, separate it from nearby bugs, and correct it by naming the constraint before adding capacity. It is grounded in Goldratt's constraint rule, Brooks's partition test, Simon on attention, and Agrawal, Gans, and Goldfarb on prediction versus decision.
Bug claim in one sentence:
The bug is the false bottleneck, where adding people, tools, or agents feels like the fix, and the real limit - a decision, a missing input, or the attention required to direct the extra capacity - stays unnamed.
What this bug is not:
A ban on hiring, automation, or parallel work. More hands are the right fix when the work is already specified, independent, and larger than one worker's time. It is also not a product review of AI-team startups, and not a claim that one agent is always enough.
Nearby bugs to check:
| Nearby bug | Difference |
|---|---|
| Law of the Instrument | Favorite method, applied too widely. False bottleneck is the earlier diagnosis: "the shortage is hands." A new AI team can be both if it is also the tool you now reach for every stuck week. |
| Action Bias | Pull toward motion rather than waiting. Capacity can be added calmly, inside a plan, with no panic. The error is which resource you increase. |
| Bikeshedding | Time goes to the topic everyone can discuss. Roles and permission screens are easy to discuss; the constraint often is not. They stack. One is about talk, this one is about what you buy. |
| Opportunity Cost Neglect | The visible price leaves out the displaced option. Extra capacity often displaces attention. This page is about misnaming the limit; that attention is usually the unnamed cost. |
| Automation Complacency | A finished-looking output ends the check. You can review every output carefully and still have built capacity you cannot feed. |
| Analysis Paralysis | You will not close a decision. Here you may close a capacity plan quickly because the real decision is harder. |
Concept gate: yes
2. Sources and research
| Source / concept | Year | What it supports | URL / DOI / note |
|---|---|---|---|
| Goldratt, E. M., & Cox, J. The Goal | 1984 | Throughput is set by the bottleneck. Speeding up other steps stacks work in front of it and does not raise output. | North River Press. No lab effect size. Do not attribute the later formal five-step list to the novel. |
| Brooks, F. P. The Mythical Man-Month | 1975 | People and months are interchangeable only when work partitions with no communication. Months follow sequential constraints; useful headcount follows independent subtasks. The "makes it later" line is his oversimplification. | Addison-Wesley. |
| Simon, H. A. Designing Organizations for an Information-Rich World | 1971 | Abundant information makes attention scarce. A new system saves attention only if it absorbs more than it sends back. | In M. Greenberger (Ed.), Computers, Communications, and the Public Interest. Johns Hopkins Press. |
| Agrawal, Gans, & Goldfarb. Artificial Intelligence: The Ambiguous Labor Market Impact of Automating Prediction | 2019 | Machine learning is a prediction technology and an input to decisions. Sometimes the decision is automated. Sometimes a person still judges the action. | Journal of Economic Perspectives, 33(2), 31-50. https://doi.org/10.1257/jep.33.2.31 |
Do not invent titles, years, DOIs, or effect sizes. No percentages.
3. Article plan
Explained boundary:
State the partition test early. Parallel, already-specified work (tickets, leads, a pile of similar files) is the case where more hands are correct. One task you still have to think about is one pass. Spare capacity that you then invent tasks for is part of the bug. Say plainly that the name is a catalog handle, not a clinical construct.
Examples planned (4-8 quote examples):
- Hire because the week is slow.
- Stand up an agent team and assign roles.
- Capacity first, tasks later.
- Add engineers while the spec is still fuzzy.
- One fast research pass, imagined as five agents.
- More publishing channels while conversion is the leak.
- Buy the tool suite so the work has somewhere to go.
- Another course before the first ugly version.
Real-world scenarios, if useful:
The ready team. The hire who is waiting on your unmade call. Research that returns in minutes and still needs a judgment. A course bought instead of a scope decision.
Impact:
Briefing and review fill the day. Invented tasks feed idle capacity. Unsplittable work gains handoffs and loses pace.
Causes:
Capacity is for sale and a decision is not. The easy question ("what can I add?") replaces the hard one ("what is the limit?"). Older weeks when hands really were short keep the diagnosis alive.
Self-spotting signs, if useful:
Can describe the capacity, cannot name the constraint. New help is blocked on you. Setup took longer than the job. No week of independent tasks to hand over. Idle capacity feels like a debt.
Prevention moves (3-6):
Name the constraint as a thing you could point at. List tasks that can run without talking to each other. If that list is short, do the constraint. If it is long, add hands. Ask whether the helper absorbs attention or emits approvals and merges.
Questions & Answers, if genuinely sharp:
When is a team the right size? Does hiding the agents and asking only for an outcome remove the bug? What if the limit really is hands?
Reframing examples (2-3):
| Label | Original thought | Reframed thought |
|---|---|---|
| The team | This research needs a team of agents, one for each role. | This is one pass, then a judgment. A team is for a pile of separate jobs I can already list. |
| The hire | We're slow, so we should hire. | Slow at an unmade call. A new person waits on that call too. |
| The channels | We need more places to publish. | Publishing is already cheap. The limit is whether people stay. |
Related thinking bugs:
| Bug | Difference note |
|---|---|
| Law of the Instrument | Favorite tool versus the prior claim that the shortage is hands. |
| Action Bias | Urge to move versus which resource you increase. |
| Bikeshedding | Discussable topic versus what you buy. |
| Opportunity Cost Neglect | Unpriced alternative versus misnamed limit. |
| Automation Complacency | Stopped check versus capacity you cannot feed. |
4. Rewrite inventory (only for existing pages)
- Current sections: new page
- Current examples worth preserving:
- Current sources:
- Current reframes:
- Current related links:
- What can be removed, and why:
5. Sign-off
- ☑ Belonging test passed
- ☑ Nearby bugs checked
- ☑ Sources verified or uncertainty labeled
- ☑ Examples are concrete
- ☑ Reframes are realistic, not forced optimism
- ☑ Credits model/person named