Where the Work Gets Stuck

The federal government can publish a movement playbook, an engineering team can identify $150,000 in annual savings, and a building-control system can expose every temperature and power reading in the place. Great. Who calls the hauler, gets three contractor bids, verifies the new assembly, or notices that a fan has been limping toward failure for six weeks?

We have become extremely good at producing signals. Reports, dashboards, sensors, recommendations, policy changes, analytics. Then somebody exports the result to a PDF and emails it to the person who was already too busy to act on it. Progress, apparently, has a paperclip icon.

That handoff is the opening. AI helps where the inputs are messy: reading reports, matching records, explaining anomalies, drafting scopes, and preparing a human decision. Deterministic software should own dates, thresholds, calculations, permissions, and audit trails. The person with the license, authority, or wrench still makes the consequential call.

Quick compare: five execution businesses

Opportunity 1: Screwworm Movement Continuity Board

The problem. When a livestock premises falls inside a New World screwworm zone, moving animals can require inspections, treatments, certificates, coordination with veterinarians, destination approvals, and timelines that vary by animal and movement type. Ranchers, livestock markets, veterinarians, haulers, and state officials may all touch the same movement. Their shared system is often a phone tree with attachments.

That hurts quickly. A missed inspection window can delay a sale or shipment. A market may need a temporary aggregation point. A veterinarian needs current rules, not the version someone saved three weeks ago. The underlying veterinary judgment is not the software opportunity. Keeping every party on the same version of the plan is.

Why now. USDA updated its national continuity-of-business movement guidance on October 6, explicitly allowing states to authorize movement-requirement activities at additional premises types, including temporary aggregation points. Texas already has active infested zones and zone-specific movement information. What is known: the response now has live geography, documents, and multi-party movement rules. My read: operational variation will grow as states apply the framework. The bet: livestock organizations will pay for a neutral board that turns the current rules into a trackable movement plan.

Existing solutions. USDA and the Texas Animal Health Commission publish the official rules and maps. GlobalVetLink helps veterinarians create animal health certificates. Those are valuable pieces. The remaining gap is continuity across the entire movement: origin, inspection, treatment, certificate, carrier, destination, deadlines, exceptions, and the exact rule version used.

The AI advantage. AI can turn updated guidance into draft checklists, extract premises and animal details from documents, summarize exceptions, and identify missing records. Geofencing, date math, role permissions, and completion status should be deterministic. A model should never diagnose screwworm, clear an animal, or invent a movement requirement.

Competition moat. The durable asset is not the checklist. It is the local workflow map: which office accepts what, who is available after hours, which documents cause rework, and how the state applies each revision. Trusted relationships with markets, veterinarians, and haulers matter more than a clever chat box.

Risk. This is disease-response infrastructure. Bad information can create animal-welfare, financial, and biosecurity harm. The product must clearly separate official requirements from draft guidance, log every source revision, and require veterinary or regulatory signoff where appropriate.

Signals: USDA’s October 6 response update, the current federal status page, and Texas movement and zone guidance.

Opportunity 2: ITAC Recommendation-to-Install Desk

The problem. Small and midsize manufacturers can receive an excellent energy assessment and still do nothing with it. The report lands on a plant manager’s desk beside production targets, maintenance emergencies, capital requests, vendor quotes, and the general industrial preference for not shutting down equipment that currently works.

Oak Ridge National Laboratory analyzed more than 20,000 assessments and 151,000 recommendations from the federal Industrial Assessment Centers program. The implementation rate was 47%. That does not mean the recommendations were bad. It means a recommendation and an installed project are different products.

Why now. On October 2, DOE opened the next five-year ITAC competition. The refreshed program will continue no-cost assessments, add a mining and mineral-processing track, and feed a network that already advertises implementation grants of up to $300,000 per qualified recommendation. DOE says a typical assessment identifies about $150,000 in annual savings. What is known: more assessments and recommendations are coming. My read: the persistent constraint is owner-side execution. The bet: factories will pay a practical project desk to convert high-confidence recommendations into bids, grant files, shutdown plans, and verified savings.

Existing solutions. ITACs deliver the assessments. Energy-service firms such as Cascade Energy offer sophisticated energy-management programs. Mechanical, electrical, and controls contractors install the work. The gap is a vendor-neutral owner’s rep for smaller plants that are not buying a large enterprise engagement but still need someone to carry the recommendation across purchasing, engineering, incentives, scheduling, and measurement.

The AI advantage. AI can parse the assessment, separate no-cost operational changes from capital projects, draft scopes of work, normalize contractor bids, assemble grant evidence, and maintain the decision log. Deterministic calculations should preserve the engineer’s baseline, payback assumptions, and measurement method. Licensed professionals approve designs. Plant leadership approves shutdowns and capital.

Competition moat. Build a recommendation-to-scope library, real installed-cost benchmarks, contractor response data, and verified savings by equipment type. Distribution through assessors and trusted local trades is harder to copy than the software.

Risk. Projects can require licensed engineering, procurement rules, utility coordination, safety review, and production shutdowns. Stay vendor-neutral, disclose referral economics, and do not turn a screening model into an engineering stamp.

Signals: DOE’s October 2 program announcement, the current ITAC program page, and ORNL’s implementation analysis.

Opportunity 3: Chassis-Free Manufactured Housing Pilot Pack

The problem. Removing a design constraint from a building code does not magically produce a financeable, inspectable, transportable, warranty-ready product. Manufacturers still need engineering details, quality-control checkpoints, transport and set plans, installer instructions, jurisdictional coordination, lender and insurer explanations, and a defensible record of what changed.

Why now. HUD’s October 7 Next Generation Manufactured Housing Action Plan follows the removal of the permanent-chassis requirement and explicitly points toward new factory-built designs. That opens real design space, but early adopters will carry a coordination tax. What is known: HUD is modernizing the national framework. My read: factories will move carefully because warranty, transport, installation, and inspection failures are expensive. The bet: a repeatable pilot pack can make the first project legible to every party without pretending the rules are finished.

Existing solutions. Offsight manages factory production, schedules, documents, and quality workflows. FTQ360 offers factory-built housing inspection and quality-control software. Those products serve ongoing operations. The opening is a narrower transition package for the first chassis-free design: requirements map, pilot gates, evidence package, field-install feedback, and a reusable path to the next model.

The AI advantage. AI can compare standards, drawings, work instructions, inspection records, and field notes; draft requirement matrices; and cluster recurring defects. Computer vision may help classify documented conditions, but it should not replace required inspections. Deterministic revision control and signed approvals matter more than fluent prose.

Competition moat. The moat is the approved path: factory-specific stage gates, inspection evidence, install feedback, lender and insurer objections, and a network of engineers and inspectors who know the new assembly. The first clean pilot becomes the template for the next ten.

Risk. Guidance will evolve. Structural design, transport, installation, and code compliance require qualified professionals. This should be a collaboration and evidence layer, not a substitute for an engineer, inspection agency, or authority having jurisdiction.

Signals: HUD’s October 7 action-plan announcement, Offsight’s factory-management platform, and FTQ360’s factory-built housing quality platform.

Opportunity 4: Local Controls Read-Only Copilot

The problem. Small commercial buildings and light-industrial sites have plenty of data and not enough attention. HVAC controllers, meters, inverters, pumps, and refrigeration systems expose temperatures, setpoints, alarms, and run states. The owner usually sees them only after comfort complaints, a high bill, or a failure. Full building-automation replacements are expensive. Cloud monitoring creates another credentialed connection into equipment that may stay in service for twenty years.

Why now. Home Assistant’s September release made shared Modbus connections easier for device integrations, and the October 7 release improved automation traces, alerts, energy views, and template climate entities. OpenRemote also documents local Modbus gateways with history and rules. What is known: local-first components can now collect and present common controls data with less custom plumbing. My read: that makes a contractor-installed, read-only baseline product practical. The bet: owners will buy useful detection without giving an AI write access to the building.

Existing solutions. Large BAS vendors, energy-management platforms, and controls contractors serve sophisticated facilities. Open-source systems such as Home Assistant, OpenRemote, and BEMOSS provide flexible local foundations. The gap is productization: safe protocol discovery, documented register maps, contractor-friendly deployment, baseline learning, human-readable findings, and an escalation path to the technician who can fix the problem.

The AI advantage. Local models can summarize trends, group recurring alarms, explain deviations from a site baseline, and turn equipment history into a technician brief. Rules should detect hard thresholds, simultaneous heating and cooling, excessive cycling, and missing data. Begin read-only. Recommendations require approval. Writes remain disabled until a controls professional deliberately commissions them.

Competition moat. Build a library of validated device profiles, site-type baselines, technician feedback, and proven fault signatures. Local deployment and contractor distribution create trust. The moat is knowing which alert deserves a truck roll and which one deserves silence.

Risk. Controls networks can be fragile and safety-critical. Protocol support is inconsistent, cybersecurity matters, and noisy alerts kill adoption. Read-only by default is not a slogan here. It is the boundary that makes the first product sane.

Signals: Home Assistant 2026.10, the September shared-Modbus release, OpenRemote’s Modbus gateway documentation, and the BEMOSS open building-management project.

Opportunity 5: Back-Catalog Sponsor Inventory Desk

The problem. A creator may have hundreds of evergreen videos that still generate views, but brand sales are organized around the next upload. Old integrations age badly, new sponsors cannot easily buy relevant back-catalog attention, and creators lack a clean inventory of which videos fit a campaign, what claims are safe, and what performance the inserted segment produced.

Why now. On September 23, YouTube said it was expanding its 2026 dynamic brand segment pilot. Invited creators can insert a creator-led segment into as many as 20 long-form videos at once, target by country, replace or remove it later, and preserve the original video’s metrics. What is known: YouTube is turning published videos into replaceable sponsor inventory. My read: the platform solved the insertion mechanic, not the catalog sales operation. The bet: mid-sized creators and managers will pay for catalog mapping, package creation, compliance review, and campaign operations.

Existing solutions. YouTube Studio handles the segment. Companies such as Agentio help creators connect with brands, while Spotter finances creator catalogs. The remaining gap is a service-first inventory desk that understands the back catalog at the video, topic, audience, geography, claim, and available-slot level.

The AI advantage. AI can index transcripts, classify evergreen topics, flag likely brand-safety and claim conflicts, suggest campaign bundles, and draft sponsor-specific talking points. Deterministic rules enforce segment length, eligibility, dates, countries, disclosures, exclusions, and approval status. The creator records the segment. The system organizes and sells the inventory.

Competition moat. Performance by topic, campaign, geography, segment style, and catalog age can become a useful proprietary dataset. Creator trust, brand relationships, and clean rights records matter. A pile of embeddings does not close a sponsorship.

Risk. The feature is invite-only, device support is still rolling out, and the platform can change access or economics. Creator-led requirements limit pure ad reuse. Start as a service for eligible channels and avoid building a broad marketplace before the pilot proves repeat demand.

Signals: YouTube’s September 23 monetization announcement and the current dynamic brand segment pilot requirements.

Where I would place the bets

• Single best opportunity: The Screwworm Movement Continuity Board.
• Highest-probability business: The ITAC Recommendation-to-Install Desk. The savings have already been identified, the customer already has the report, and a service can earn revenue before the software is finished.
• Highest-upside business: The Chassis-Free Manufactured Housing Pilot Pack. A successful first pilot can become reusable infrastructure for a much larger factory-built housing shift.
• Fastest path to first revenue: The Screwworm Movement Continuity Board. Sell managed coordination around the next five affected movements.
• Most overlooked: The Local Controls Read-Only Copilot. Better open tools make a safe, contractor-led product possible without replacing the entire controls stack.
• Best future That SI Guy article: The Chassis-Free Manufactured Housing Pilot Pack. Removing one steel constraint from a fifty-year housing system creates a surprising number of downstream questions.
• Best standalone-company candidate: The Screwworm Movement Continuity Board, expanded into livestock movement continuity for disease events, weather disruptions, and other regulated exceptions.

The next move

Interview one Texas livestock market or regional hauler about the last five movements affected by screwworm rules. Reconstruct each movement on one sheet: people involved, calls made, documents exchanged, deadlines, delays, and the moment somebody had to ask, “Are we actually cleared to go?”

Then offer to run the next five as a fixed-price coordination desk. No platform pitch. No synthetic demo. Just fewer missed steps and one current answer everyone can see.

The playbook exists. Build the part that gets the cattle moving.