Research · 2026 to 2036
Tomorrow's founders, tested today.
How should founders and companies work with AI agents? Every summer we test it with 350 startup teams and count verified revenue, not opinions.
- randomised teams a year
- 350
- sites: London, Rome, Lisbon, Lausanne
- 4
- revenue target in 30 days
- $10K
- revenue, not reported
- Verified
HostsUC Berkeley · University of Lisbon · Bayes Business School · Carnegie Mellon University
PartnersMicrosoft Research · Google DeepMind · OpenAI · NVIDIA · Y Combinator · Antler · Hugging Face
2026 to 2036
Six questions for the next ten years
AI moved into startups and companies fast in 2025 and 2026. Here is what the latest data says, and the EIA study that tests each question.
Solo founders
Now
36.3% of new startups on Carta had a single founder in the first half of 2025, up from 23.7% in 2019.
Agents at work
Now
23% of companies say they are scaling an AI agent. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027.
McKinsey, The state of AI in 2025 (opens in a new tab) · Gartner, 25 June 2025 (opens in a new tab)
Open question
How much can a founder hand to agents while the money is real, and who stays in charge?
Accelerator studyCompany adoption
Now
20% of EU companies with 10 or more staff used AI in 2025, up from 13.5% a year earlier.
Open question
Where in a small company does AI pay back, and where does it only add cost?
Finding: where AI sitsSame ideas
Now
Stories written with one AI idea were 10.7% more alike than stories written without AI.
Health AI
Now
Over 80% of US physicians now use AI at work, double the share in 2023.
AMA, Physician Survey on AI, March 2026 (opens in a new tab)
Open question
Can a model test a health tech startup before the founder spends a year building it?
World HealthTech ModelFirst revenue
Now
20% of startups incorporated through Stripe Atlas charged their first customer within 30 days in 2025, up from 8% in 2020.
Stripe, annual update, 24 February 2026 (opens in a new tab)
Open question
Can we tell before a company is founded whether it will send its first invoice within 30 days?
FounderBench study
Five studies
One cohort, five questions
Every summer 350 teams start companies at EIA. Each study uses the same cohort and the same verified revenue.
Course
- When
- Jul–Aug, 30 days
- Sample
- 350 teams, randomised
- Question
- Can a founder-controlled agent system reach $10K revenue in 30 days, and which control setup keeps the founder in charge?
- Design
- 350 teams, randomised in three strata. Solo (100): CEO plus product and GTM agents. Pair (200): CEO and CPO, AI as CMO, 2×2 oversight by self-improvement. Trio (50): human CMO owns GTM.
- Endpoint
- Verified revenue at day 30. Power d≈0.57 per 50-team arm.
World HealthTech Model
A world model of how startups grow
Trained on real EIA cohorts and checked against their verified revenue. Health tech is the first field we are building it for.
Watch the model run a startup before the founder does.
How it works
- 01State encoderOpen LLM plus structured features
- 02Dynamics corePhysics-informed
- 03EvaluatorCalibrated on verified revenue
- 04Control policyWhen to ask the founder
Roadmap
- 2026Control layer and trace schema
- 2027World model v0.1 beats mentor scores
- 2028FounderSim reproduces the effect sizes
- 2029Planning agents beat agents without a model
- 2030Adopted by 5+ outside teams; 6+ peer-reviewed papers
Technical detail: subsystems and calibration data
Four subsystems
Demand
- Structure
- Diffusion ODEs, saturation, price elasticity
- Learned
- Neural ODE with residual loss
- Priors
- Product Hunt, Amazon Reviews, Census entry series
Product
- Structure
- Quality-adoption coupling; PDE feasibility for hardware
- Learned
- Neural operators, video world models
- Priors
- The Well, PDEBench, automotive CFD sets
Go-to-market
- Structure
- Funnel conservation, capacity queueing, channel saturation
- Learned
- Constrained sequence model
- Priors
- YC and Antler descriptions, Product Hunt metadata
Business
- Structure
- Cash conservation; MRR = acquisition minus churn
- Learned
- Hybrid grey-box model
- Priors
- Open Startups and Indie Hackers revenue series
Calibration data
- firms in the Y Combinator directory (CC0)
- 5,404
- firms in the Antler portfolio
- ~1,600
- rows of PDL company data
- 32.3M
- VCBench profiles
- 9,000
- human choices in Psych-101
- 10M+
- physics corpus in The Well
- ~15 TB
Findings
What the research says so far
Seven findings from the literature that the studies build on.
- 01
Solo plus AI matches a team
With AI access, solo founders match team outcomes.
- Dell'Acqua et al., "The Cybernetic Teammate," Organization Science
- Kim et al., "Generative AI Fuels Solo Entrepreneurship," arXiv:2605.10291 (opens in a new tab)
- 02
Value depends on where AI sits
Teaching 515 startups where AI fits raised revenue 1.9× and cut capital needs 39.5%.
- Kim, working paper
- 03
Long-horizon autonomy is unsolved
Agents derail in Vending-Bench. 3 of 12 models stay profitable in YC-Bench.
- 04
Revenue is not a clean reward
Self-improvement works on benchmarks, but evaluators must co-evolve to stop reward gaming.
- Zhang, peer-reviewed
- 05
Human control is untested with real money
L1–L5 autonomy and oversight regimes are formalised, but never run with founders and live revenue.
- 06
AI ideation homogenises
New-to-the-world creation needs measurement and de-fixation.
- Wenger & Kenett, "Large language models are homogeneously creative," PNAS Nexus 5(3)
- 07
Physics can constrain business dynamics
Neural operators embed governing equations; neuro-symbolic frameworks impose economic constraints.
- Ahmadi et al., "Physics-Informed Machine Learning in Biomedical Science and Engineering," Annual Review of Biomedical Engineering
- "ARTEMIS," arXiv:2603.18107 (opens in a new tab)
- "PhyWorld," preprint
Published from EIA cohorts
Earlier studies of EIA students.
Founding intent rose from 0% to 58% among students with no prior startup experience.
Taggar, Domurath & Coviello, Academy of Management, 2019
Peer-driven learning raised entrepreneurial passion by 1.3 points on a 5-point scale.
Taggar, Domurath & Coviello, 2024
Self-efficacy rose by 1.8 and resilience by 1.6 on a −3…+3 scale.
Brandt, 2023
Join a study
Who are you?
Pick what you can bring. We reply within one business day.
Since 2013
A research partner with thirteen years of cohorts
The mission passes from cohort to cohort, like a torch. 25,000 carriers so far.
Photo credits
City photos: Wikimedia Commons contributors (CC BY / CC BY-SA), listed on the About page. The rest are EIA's own.