Idea Lab
Pick a vertical. The radar reads everything it tracks there - momentum by year, crowded subniches, dead companies, fresh rounds, grant money, open YC RFS themes - and proposes concrete startups someone should build. The companies below are fictional; the data behind them is not.
Robotics and physical world
296
companies, last 2 years
4052
tracked all-time
88%
of checked sites alive
$2259M
across 4 recent rounds
Most crowded right now: data centers (5)
concept 1
Sumpline
Monthly microbial and fluid-health testing for machining coolant sumps, with an action sheet per sump.
A subscription testing service for metalworking fluid: each month a plant sends one 20 ml sample per coolant sump in a pre-labelled kit, Ravi runs a rapid quantitative microbial and pH/tramp-oil assay, and the plant gets a one-page trend sheet per sump saying skim, dose, or dump. Buyers are maintenance and production managers in machining and stamping plants who currently guess, or rely on dip slides read by eye. Later Marta machines an inline sampling port and sampling jig so shift staff take repeatable samples in 30 seconds.
- why now
- Robotics and physical world shows 142 tracked companies in the last two years and 88 in the 2024 cohort alone, yet none of the 35 most recent entrants touch fluid or consumable maintenance - the money is going to robot brains and humanoids. Ember Robotics ('Datadog for robots') went inactive in 2024, which says dashboards without a physical measurement do not stick in plants. Federal manufacturing attention is live too: the NIST Manufacturing Extension Partnership Center State Competition closes 2026-08-21 and NSF has open 'Advanced Manufacturing' and 'Manufacturing Systems Integration' calls that MEP centres can co-fund pilots against.
- wedge
- Five plants, six sumps each, flat monthly fee per sump, paper report only. Ravi needs roughly three bench hours a week; Marta needs two calls and one plant walk per customer. No software, no instrument purchase, invoice from month one.
- closest real companies
- None tracked in the briefs. Nearest tracked neighbours are Grip (physical AI for waste management) and Dawn Industries (automatic diagnosis and fix for industrial automation cells) - both act on machines and software, not on the fluid chemistry that kills the machines.
- main risk
- Plants treat coolant as a consumable and accept free, low-quality testing from their fluid supplier rather than paying an independent third party.
concept 2
Benchcell
Single-purpose bench robots for the one step a core facility repeats a thousand times a month.
Build-to-order, single-task bench machines for academic core facilities: plate reformatting, cap and seal handling, or loading a specific instrument, priced at a level a facility can put on a purchase order without a capital committee. Each unit is a hard-tooled fixture with a small motion axis and PLC-grade control, no cloud, no app, documented so the facility's own technician can service it. Buyers are core facility managers and platform heads who have queues but no automation budget for a six-figure liquid handler.
- why now
- The 2026 robotics cohort is 77 companies and the YC Fall 2026 request is literally 'New Operating Systems for the Physical World', but the recent list runs to general-purpose bets - Nori's sub-$2000 humanoid, Shiraz AI, Neuromorphic - while K-Scale Labs (open-source humanoids) went inactive in 2024. Nobody in the 142 recent robotics companies is selling a boring, single-purpose bench fixture into labs, and 84% website survival in this cluster shows hardware with a defined job survives.
- wedge
- One machine, one step, built for Ravi's facility first using existing grant or service budget, then quote the same design to three peer facilities at a fixed price with a four-week lead time. Cap output at roughly one unit a month so it never becomes manufacturing scale-up.
- closest real companies
- None tracked in either brief for lab bench automation. twentyfour26 (on-demand CNC prototyping) and Ultrasonium sell parts, not validated task machines; Manifold and OS3 sell robotic labour into warehouses, a different buyer and price point.
- main risk
- Academic purchase orders take longer than the eight-month runway and the first three quotes stall in institutional procurement.
concept 3
Lotproof
Independent lot-performance testing that tells labs and vendors which reagent batch will actually work.
A fixed-protocol testing service that scores incoming reagent and enzyme lots on real assay performance, not on the certificate of analysis, and returns a comparable score plus a trend across lots. Sold two ways: to reagent vendors and distributors who want third-party evidence their batches are consistent, and to labs and facilities that want to screen a lot before committing a project to it. Marta's fixed test jig keeps the per-lot cost low enough to be routine.
- why now
- Attestation of physical-product quality is being funded right now - Complir (YC X26) is pitched as 'Vanta for physical products' - while classic B2B SaaS cohorts collapsed from 179 new companies in 2021 to 38 in 2026, so the shape that works here is a measurement service, not a dashboard. Nothing equivalent exists for reagent lots in either brief, and batch inconsistency is a failure Ravi already watches labs hit repeatedly.
- wedge
- Publish, from anonymised existing data, a short note showing how much assay variance traces to lot rather than protocol, then offer three reagent vendors in the network a paid ten-lot retrospective test at a fixed price. That is one deliverable, sellable inside two months.
- closest real companies
- Complir (Vanta for physical products) is the closest tracked shape but targets product compliance, not reagent lot performance; Shielded (supply chain and financial risk) sits on paperwork rather than bench measurement. No direct competitor tracked.
- main risk
- Vendors refuse to fund data that could prove their own lots inconsistent, and labs expect the vendor to pay for it.
concept 4
Jigsmith
Two-week custom lab hardware: adapters, carriers and jigs quoted from a photo and an assay description.
A small-batch machining service for the one-off metal and plastic parts labs cannot buy: instrument-specific plate carriers, tube racks with the right pitch, alignment jigs, anti-contamination workstations, retrofit brackets for a discontinued instrument. Quoting happens from a photo plus a sentence about what the part must do in the workflow, and delivery is two to three weeks. Buyers are core facility managers, platform technicians and PIs with equipment money and no engineer.
- why now
- On-demand machining is being funded - twentyfour26 (YC F26) does on-demand CNC prototyping and Ultrasonium (YC S26) is scaling metal parts - which proves demand but leaves the low-quantity, high-context lab jobs unserved because they are too small for those shops. Meanwhile the graveyard shows the software-only route into hardware buyers is thin: Anchorhead (fractional buyer for hardware companies) and Smartbase (AI-native ERP for manufacturers) both went inactive in 2026.
- wedge
- Ten fixed-price parts for Ravi's own facility and the two labs next door, invoiced within weeks. Publish a one-page catalogue of the three parts that get asked for twice.
- closest real companies
- twentyfour26 (on-demand CNC prototyping) and Ultrasonium (mass metal parts) - both compete on throughput and ignore single-unit, assay-context jobs; GUILD and ProvenMetal serve defence and PCB, not benches.
- main risk
- It stays a paid hobby: revenue is capped by Marta's hours at the mill and never becomes a product.
concept 5
Coreworks
Paid instrumentation engineering for labs, written as a subaward into grants they already hold.
A two-person engineering shop that other research groups and core facilities hire, out of already-awarded instrumentation and equipment money, to design, machine and document a bespoke module: a custom sample stage, a temperature-controlled fixture, a motorised retrofit for an ageing instrument. Deliverable is a working device plus CAD, bill of materials and a maintenance sheet, so the lab is not dependent on the team. Buyers are PIs and facility heads with grant money and no mechanical engineer.
- why now
- Public money for exactly this exists and is visible in the brief: NSF awards such as the ~$188k human-robot interaction collaborative project and ~$240k collaborative perception project, the open NSF 'Manufacturing Systems Integration' and 'Advanced Manufacturing' calls, and ~$50k I-Corps translation awards. Ravi's own note that grant money, not savings, is the real runway matches a business whose revenue line is a subaward rather than a venture round - which Marta vetoes anyway.
- wedge
- Two paid subawards on grants that are already funded, sourced from Ravi's peer PIs, delivered inside a semester. Fixed scope, fixed price, no cost-plus.
- closest real companies
- None tracked. Torus (Legora for physical engineering firms) and Osseus (robotics development platform) sell tools to engineers rather than doing the engineering for labs.
- main risk
- Grant and subaward timelines run longer than the eight-month runway, so cash arrives after the shortest runway ends.
Fictional concepts generated 2026-08-18 by claude-opus-5 from MarkosWeb data (accelerator portfolios, funding news, US federal grants, YC RFS). Numbers reflect tracked sources only; treat every concept as a research prompt, not a plan.