# Future To Build: The Ideas Catalogue

Research checked September 9, 2026.

The essay now ends with full original write-ups for all 65 current ideas. This companion remains the compact reference, with commercial tests and the earlier SPC question indexes.

This companion indexes every top-level entry in the three current collections: 42 from Conviction, 13 from YC’s Fall 2026 edition, and 10 from SPC’s Summer 2026 edition. It also indexes four earlier SPC collections. It does not cover every historical YC batch.

The topic labels are concise paraphrases. The questions and business designs below are my own analysis, not claims or recommendations attributed to the source authors. The linked originals contain their full proposals, examples, and subquestions; this catalogue is not a reproduction of those texts.

## Ten opportunities I would investigate first

These are editorial priorities, not a forecast of investment returns. I favor a clear customer, a measurable improvement, and a path to learning from delivery. A technically difficult idea is not automatically a good company.

| Opportunity | First customer and starting product | Commercial hypothesis | First proof required | Reason to stop or change direction |
|---|---|---|---|---|
| Evidence for automated work | An engineering team; review packets for one class of release | Annual platform fee tied to supported workflows | Review becomes faster while escaped defects do not increase | Reviewers still reconstruct everything manually |
| Contestable organizational memory | Procurement team; a record of supplier decisions and changed assumptions | Department subscription with a paid historical setup | New staff can recover and challenge decision reasoning faster than with search | Historical opinions are mistaken for current facts |
| Shared agent authority | Customer implementation team; commitments and approvals across departments | Price per active implementation, with a minimum subscription | Fewer contradictory promises and reopened handoffs | The product adds coordination overhead without replacing any |
| Portable personal context | Households coordinating recurring obligations | Subscription paid by the person whose information is stored | Move to a second assistant while retaining useful context and access boundaries | Export works technically but is unusable elsewhere |
| Maintenance for small applications | Operations teams generating internal tools | Managed application fee, including upkeep and retirement | A tool survives owner departure and an integration change | Maintenance is bespoke consulting on every application |
| Flexible compute delivery | A buyer with batch jobs and explicit deadlines | Margin on accepted completed jobs | Lower cost after retries, fallback, and support | Savings disappear under realistic service requirements |
| Independent experiment replication | Laboratories sharing one repeatable assay | Paid verification projects, then recurring workflow access | Different operators reproduce results within an agreed tolerance | Customers value attractive results more than reliable ones |
| Recovery-centered robotics | One commercial site with a repetitive task | Service contract with an explicit intervention allowance | Interruptions can be resolved without permanent expert staffing | Remote support labor rises as quickly as deployments |
| Apprenticeship that builds judgment | Employer with a recurring training burden | Employer-paid training and assessment | Learners perform unfamiliar tasks independently | Better assisted performance masks weaker independent skill |
| Care coordination | Families and a partner service provider | Family subscription or a transparent service-provider contract | Fewer unresolved responsibilities with less chasing | The person receiving care experiences more intrusion than benefit |

The first five can be investigated through relatively contained software pilots. Compute, laboratory operations, and robotics require a different financing and operating plan. Education and care depend especially on proving that the affected person benefits, even when someone else pays.

## Conviction: all 42 topics

Source: [Conviction’s evolving startup collection](https://www.conviction.com/startups.html). Entries follow the body order, which differs slightly from the page’s quick links. C08 corresponds to the inference-grid anchor in the original request.

The right-hand column is an independent question I would put to a founder. It identifies what could break the business, not what Conviction claims has been solved.

| ID | Topic | My discriminating question |
|---|---|---|
| C01 | Laboratory throughput | Can two customers accept the same deliverable specification, or does every contract require a new operation? |
| C02 | Manufacturing know-how | If the most experienced operator leaves, does the next shift maintain quality without their informal interventions? |
| C03 | Photonics testing | Will a customer allow your test to reject an expensive component, and what independent evidence earns that trust? |
| C04 | Transformer reuse | Who accepts the warranty exposure when historical maintenance information is missing or contradictory? |
| C05 | Analog automation | Can a customer reproduce your result on its own tools, under its own assumptions, before relying on it? |
| C06 | Insurance portals | How does the service distinguish a successful submission from a screen that merely appears to confirm success? |
| C07 | Simulation calibration | When a prediction misses, can you separate instrument error, manufacturing variance, and an incorrect physical assumption? |
| C08 | Distributed inference | Does a lower site-level cost remain an advantage after accounting for duplicate work and correlated failures? |
| C09 | Regulatory intelligence | Can a reviewer trace every factual assertion to its dated source and identify where human interpretation begins? |
| C10 | Software assurance | What important failure can your evidence detect that the customer’s existing test suite would miss? |
| C11 | Robot assurance | Can you demonstrate that a new test changes an engineering decision, rather than merely producing a report? |
| C12 | Chip verification | Is a discovered issue reproducible by the customer, and does it alter a release decision before a deadline? |
| C13 | Adaptive underwriting | Are better predictions still useful when data arrives late or customers change behavior in response to measurement? |
| C14 | Organizational reasoning | Can users inspect evidence against the system’s favored explanation and correct an inherited misconception? |
| C15 | Agent tooling | Can the same operation be retried safely after an ambiguous response, without duplicating its external effect? |
| C16 | Bespoke software | Who owns the application when its original creator leaves, and how does the company retire it safely? |
| C17 | Domain execution | Which task remains valuable when the underlying model improves enough to make your current implementation trivial? |
| C18 | Assisted trades | Does the worker acquire transferable judgment, or become dependent on instructions that fail outside familiar conditions? |
| C19 | Workplace simulation | Does performance in the simulator predict an independently measured outcome at a real customer? |
| C20 | Enterprise reasoning | Can conflicting definitions be resolved by the actual business owner instead of silently reconciled by a model? |
| C21 | Task-specific robots | What fraction of service cost comes from recovery, travel, and support rather than the robot’s normal operation? |
| C22 | Preclinical prediction | What prospective result would establish usefulness without selecting only the cases where the model looks good? |
| C23 | Business app generation | Will the security owner approve applications made by a nontechnical team without personally rebuilding each one? |
| C24 | IT inventory | Can the system detect that an apparently authoritative record is stale before someone acts on it? |
| C25 | Deployment feedback | Does observed customer behavior provide a valid success signal, or merely reward the most convenient shortcut? |
| C26 | Compute architecture | Is the advantage visible on the buyer’s workload after migration and support costs, rather than a favorable benchmark? |
| C27 | Spatial generation | What downstream task requires the generated object to be physically consistent rather than simply visually convincing? |
| C28 | Data cataloguing | Can a change in meaning be detected even when the field name and data format stay the same? |
| C29 | Business acquisition | Would the operating improvement still justify the purchase if the automation took twice as long to implement? |
| C30 | Incident diagnosis | Can a wrong recommendation increase the incident’s scope, and how quickly can a responder recognize that? |
| C31 | Merchant assistance | Does the service improve contribution margin after returns and fulfillment costs, or only increase clicks? |
| C32 | Personal composition | Can users identify and reverse a mistaken inferred preference before it affects an important communication? |
| C33 | Video interpretation | Which events must never be missed, and can the buyer tolerate the false-alarm volume required to catch them? |
| C34 | Generative play | Do friends return because they created something together, or does the product depend on endless individual consumption? |
| C35 | Financial controls | Can an independent reviewer reproduce the conclusion without treating the model’s output as the evidence? |
| C36 | Purchasing assistance | Who pays the agent, and can the user inspect whether that payment changed the recommendation? |
| C37 | Legacy migration | Can the old and new systems run against the same historical cases, including rare exceptions, before a cutover? |
| C38 | Public-service access | How will the tool display conflicting or outdated official information without presenting a guessed resolution as authoritative? |
| C39 | Materials development | Does the improvement survive scale-up, variable inputs, and the customer’s actual production process? |
| C40 | Web information | Can the customer distinguish a live observation from cached information that has silently become obsolete? |
| C41 | Telephone service | Can the caller reach an accountable person after an unusual request without having to repeat the entire story? |
| C42 | Employee operations | Does one corrected employee record propagate accurately, or does the system create contradictions across connected tools? |

## Y Combinator: all 13 current requests

Source: [YC Requests for Startups, Fall 2026](https://www.ycombinator.com/rfs), the default edition on the live page when checked. Earlier editions remain selectable on that page; they are outside this current-edition index.

| ID | Topic | My discriminating question |
|---|---|---|
| Y01 | Childhood tutoring | Does progress persist on unfamiliar exercises without the tool, and can parents understand how progress was measured? |
| Y02 | Defense systems | What independent acceptance evidence establishes that a component meets its purchaser’s stated operating requirements? |
| Y03 | Small-app hosting | Can an administrator identify every application using a departing employee’s credentials and transfer ownership correctly? |
| Y04 | Collaborative agents | When two authorized colleagues disagree, can the system pause the contested action while continuing unaffected work? |
| Y05 | Offshore computing | Has the operator tested maintenance access, equipment degradation, connectivity, and downtime together under realistic conditions? |
| Y06 | Mass-market AI | Would people return if the product lost its novelty and the company stopped subsidizing expensive sessions? |
| Y07 | Older-adult support | Does the person using the product control its behavior when the purchaser is a family member or institution? |
| Y08 | Physical-work coordination | Does assigning a task to a different worker type reduce total completion cost, including exceptions and handoffs? |
| Y09 | Crypto infrastructure | Which customer problem improves compared with the existing payment or recordkeeping method, after all operational costs? |
| Y10 | Physical sensing | What decision changes because of the additional measurement, and is that change worth the collection expense? |
| Y11 | Human authentication | Does confirming identity establish authority for this particular action, or is the product confusing those two claims? |
| Y12 | Compliance operations | Can a professional reviewer distinguish applicable source material from the system’s interpretation and unresolved uncertainty? |
| Y13 | API maintenance | Does a generated migration preserve business behavior, including the cases a provider’s documentation does not describe? |

## South Park Commons: all 10 current questions

Source: [SPC’s Summer 2026 collection](https://blog.southparkcommons.com/p/requests-for-curiosity-summer-2026), dated June 17, 2026. These are exploratory questions. Scenarios about employment, medicine, or model progress should not be mistaken for established outcomes.

| ID | Topic | My discriminating question |
|---|---|---|
| S01 | Relational AI | Can a user clearly understand the system’s limits even when the interaction feels emotionally persuasive? |
| S02 | Expanded supply | Does reducing the provider’s effort create new paying demand, or merely lower prices for existing transactions? |
| S03 | Energy abundance | Which engineering constraint can be measured at the proposed customer site before substantial capital is committed? |
| S04 | Learning depth | Can the learner explain a new problem in their own terms after the product is removed? |
| S05 | Enterprise model fleets | Would several specialized models outperform a simpler system after maintenance, evaluation, and staffing are included? |
| S06 | Robotics deployment | What ordinary disruption requires a human rescue, and can that intervention be made fast and repeatable? |
| S07 | Personal memory | Can someone correct or remove a sensitive inference everywhere the system uses it, rather than only hide its display? |
| S08 | Patient agency | Does the user understand what a tool can establish, what remains uncertain, and when professional assessment is needed? |
| S09 | Ambient devices | Is local processing essential to usefulness, or an expensive feature that customers would not pay extra to retain? |
| S10 | Purpose beyond employment | Does the experience build relationships or abilities that persist outside the product, rather than require permanent dependence? |

## The earlier SPC questions, indexed

These short historical indexes preserve breadth without presenting older forecasts as current facts. They are not additional developed business plans. Labels follow each article’s order.

### November 18, 2024: the original collection

The [original 25-question collection](https://blog.southparkcommons.com/p/request-for-curiosity) covers:

1. Software lifecycle expansion.
2. Agent workforce coordination.
3. Reliability engineering assistance.
4. Alternative model architectures.
5. Persistent-inference hardware.
6. Drone development platforms.
7. Children’s creative tools.
8. Collaborative AI interfaces.
9. Established credibility providers.
10. New consumer experiences.
11. Creative mastery.
12. Media consumption.
13. Personal data ownership.
14. Agent-mediated advertising.
15. Machine-oriented interaction.
16. Agent communication protocols.
17. Faster human-computer interaction.
18. Independent business ownership.
19. Global and domestic priorities.
20. Simulation-led development.
21. Indirect AI beneficiaries.
22. Commerce discovery.
23. Compliance and consumer protection.
24. Indian financial technology.
25. Meaning and leisure.

### December 19, 2024: member questions

The [16-question member edition](https://blog.southparkcommons.com/p/member-request-for-curiosity) adds:

1. Alternatives to language-model intelligence.
2. Lifelong agent relationships.
3. Specialized service businesses.
4. Social-network economics.
5. Email-based experiences.
6. Deliberative reasoning structures.
7. Incentives for original data.
8. Tiny organizations at scale.
9. Advertiser-controlled intelligence.
10. Learning after advanced AI.
11. Knowledge-work interfaces.
12. Public-worker assistance.
13. Compute-intensive search.
14. User-directed content consumption.
15. Technology supporting human experience.
16. Aging and end-of-life needs.

### July 2, 2025: summer questions

The [17-question Summer 2025 edition](https://blog.southparkcommons.com/p/spc-requests-for-curiosity-summer-2025) covers:

1. Agent-mediated professional relationships.
2. Changing career paths.
3. Longevity-oriented care.
4. Personalized schooling.
5. Adaptive interfaces.
6. Accelerated discovery.
7. Consumer electronics.
8. Inbox intelligence.
9. Participatory entertainment.
10. Creator identity.
11. Human connection.
12. Shared organizational AI.
13. Model-centered programming.
14. Alternative robot forms.
15. Indian industrial technology.
16. Business defensibility.
17. Agency under abundant intelligence.

### December 3, 2025: winter questions

The [12-question Winter 2025 edition](https://blog.southparkcommons.com/p/spc-requests-for-curiosity-winter-2025) covers:

1. Scientific communication.
2. New monetization models.
3. Attention and entertainment formats.
4. Broader AI accessibility.
5. Compute sustainability.
6. Continuous distributed learning.
7. Expanded security exposure.
8. Programmable physical systems.
9. New measurement capabilities.
10. Stronger interpersonal relationships.
11. Agent-era governance.
12. Industrial training institutions.

## The decision framework

Before treating any of these as a company, write down five things in ordinary language:

1. **The unfinished job.** What is someone currently unable to complete well?
2. **The buyer.** Who has both the problem and authority to pay for a solution?
3. **The complete cost.** Include supervision, failures, maintenance, distribution, and physical operations where applicable.
4. **The disconfirming result.** What observation would make you abandon or materially change the idea?
5. **The customer’s retained capability.** If they leave, what knowledge, records, or abilities can they take with them?

A useful pilot compares the current process against the proposed one on real work, with acceptance conditions agreed in advance. No universal success-rate threshold fits all these domains. A game, an industrial component, and a care service have different consequences when something fails.

The best starting point is where you can obtain honest evidence quickly enough to change your mind.