Y Combinator’s Spring 2026 Request for Startups (RFS) on AI guidance for physical work is aimed at products that help people carry out real-world tasks with greater speed, safety, consistency, and confidence. The opportunity spans factories, warehouses, farms, construction sites, repair operations, healthcare, and other environments where work happens through hands, tools, machines, and movement.
For Indian founders, this is a practical category with a large potential user base—but it is also a demanding one. A convincing company will need more than a general-purpose chatbot or a computer-vision demo. It must solve a costly operational problem, work in imperfect conditions, and show why workers and managers will adopt it.
What the YC request is really looking for
“Guidance” can take several forms. An AI system might observe a task through cameras, understand a worker’s spoken instruction, retrieve the right procedure, flag a safety risk, or provide step-by-step feedback through a phone, headset, tablet, or wearable. The strongest products connect these capabilities to a measurable operational outcome.
Potential examples include:
- Assembly and quality: guiding operators through variants, checking component placement, and catching defects before dispatch.
- Maintenance and repair: helping technicians diagnose equipment, identify parts, and follow approved procedures.
- Warehousing and logistics: directing picking, packing, loading, and inventory exception handling.
- Construction: supporting site inspections, installation sequences, safety checks, and documentation.
- Agriculture: assisting with crop inspection, spraying decisions, equipment operation, and post-harvest handling.
- Healthcare and care work: guiding repetitive procedures, recording observations, and reducing documentation burden.
This does not require full humanoid robotics. In many settings, a reliable visual assistant, voice interface, or workflow layer can create value sooner than an autonomous machine. Founders building deeper robotics infrastructure may also study open-source robotic operating system frameworks before choosing their hardware and software architecture.
Why this category is difficult
Physical work creates constraints that do not appear in a browser-based workflow. A model may face poor lighting, dust, noise, intermittent connectivity, changing tools, regional languages, protective equipment, and workers who cannot stop to type. A recommendation that is merely inconvenient in an office can be dangerous on a factory floor.
Your product should therefore answer five questions early:
1. Who is the user? Distinguish the operator, supervisor, safety officer, trainer, and enterprise buyer.
2. What decision or action improves? Define the task precisely rather than describing a broad industry.
3. What information does the system need? This may include video, audio, sensor data, manuals, work orders, or machine telemetry.
4. What happens when confidence is low? Build escalation to a human instead of forcing an answer.
5. How will value be measured? Use metrics such as reduced rework, shorter training time, fewer incidents, higher throughput, or improved first-time fix rate.
Safety and permissions must be designed from the beginning. Teams handling cameras, worker voices, location data, or performance records should define retention, access controls, consent, and audit trails. The principles in this guide to securing autonomous AI workflows are especially relevant when an AI system can trigger actions or influence safety-critical decisions.
Product patterns worth testing
1. Multimodal work assistants
A worker can ask a question by voice, point a camera at a component, and receive a short instruction. This is useful where hands are occupied or literacy and language vary. A practical first version may combine speech recognition, retrieval from approved manuals, image understanding, and human escalation. Review how voice agents work before assuming a conversational interface is sufficient for noisy environments.
2. Procedure and training copilots
Many Indian employers repeatedly train frontline workers because of attrition, expansion, or process changes. An AI copilot can turn standard operating procedures into task-specific guidance, quiz workers, and identify where training breaks down. The product must preserve the approved procedure and make revisions traceable; improvisation is not a feature in a regulated or hazardous workflow.
3. Vision-based quality and safety systems
Computer vision can identify missing components, unsafe zones, incorrect handling, or deviations from a work sequence. Start narrowly. A model that reliably checks one high-volume operation is more valuable than a broad safety platform that generates untrusted alerts.
4. Human-robot collaboration
Robots can handle predictable movements while AI helps workers configure, supervise, or recover from exceptions. This can be attractive in small and mid-sized Indian factories that cannot justify a fully automated line. The key product question is not whether the robot is impressive, but whether deployment reduces total operating cost without creating new training or maintenance burdens.
Building for Indian operating conditions
India offers a strong test market because operations often combine modern equipment with legacy machines, paper-based processes, contractor workforces, and multiple languages. Design around these realities rather than treating them as edge cases.
- Support Hindi and regional languages where workers need them, while keeping technical terms accurate.
- Plan for offline or low-bandwidth operation, with safe synchronisation when connectivity returns.
- Make hardware resilient to heat, dust, vibration, and frequent charging constraints.
- Use short audio, visual cues, and large controls instead of dense dashboards.
- Integrate with existing ERP, maintenance, warehouse, or work-order systems only where integration changes the outcome.
- Establish whether the buyer is a plant head, contractor, enterprise IT team, or equipment manufacturer.
A prototype can begin with a phone, industrial tablet, body camera, or headset. Do not purchase specialised hardware before proving that the guidance loop improves a real task. Teams building the intelligence layer can also evaluate AI agent frameworks for developers in India, but the framework should remain subordinate to reliability, latency, and deployment constraints.
What to show in a YC application
YC will expect a clear explanation of the problem, product, market, and evidence. For this RFS, include concrete operational detail:
- The exact worker and task you serve.
- Why current training, manuals, supervisors, or automation are insufficient.
- A short demo showing the system in a real or realistic environment.
- Baseline and post-deployment metrics, even from a small pilot.
- Who pays, how much the problem costs, and how procurement works.
- What data advantage or workflow integration becomes stronger with use.
- How the system handles uncertainty, privacy, safety, and human override.
A strong application does not claim that AI will replace all manual labour. It shows how a specific worker can complete a valuable task more reliably and how the company can expand from one workflow to adjacent sites, machines, or industries.
A practical validation plan
Start with ten to twenty users performing one repeated task. Observe the workflow in person. Record failure modes, not just successful demos. Compare the AI-assisted process with the existing method across time, errors, rework, training, and user acceptance.
Then run a limited paid pilot if possible. A paying customer is stronger evidence than a large number of free trials, particularly in enterprise and industrial markets. Document deployment time, hardware requirements, support effort, and the person responsible for approving rollout. These details reveal whether the product can scale beyond a founder-led pilot.
The best candidates for the Spring 2026 RFS will combine a sharp operational wedge with ambitious technology. Build for the worker who needs dependable help in the moment, the manager who needs measurable improvement, and the buyer who needs a safe path from pilot to production.