Ndea

June 17, 2026

Public disclaimer: This is an independent public teardown based solely on publicly available information retrieved on 2026-06-17. It is research support, not investment, legal, tax, or financial advice. Company-provided statements are labeled as claims, not verified facts. Ndea did not participate in or review this report. Errors and omissions are possible; final decisions remain with the reader.

François Chollet wrote Keras and the ARC-AGI benchmark the field still cannot beat. Now he and a Zapier co-founder have set out to reach artificial general intelligence through program synthesis. The pedigree holds up on the public record — the open-source footprint, the Y Combinator profile, and the named seed backers all check out. Seventeen months on, what does not show up anywhere is a product, a price, a reproducible Ndea-authored result, or even a confirmed funding figure. Across a single day, five partners worked through 43 public sources, every claim cited and none lifted from a deck. The whole case rests on one unproven thing — whether a frontier breakthrough here ever becomes a company an investor can own, or just becomes open science the rest of the field copies for free. The teardown comes back with the answer. Decide before reputation alone prices the round.

What The Company Does

Ndea is a frontier AI research lab that says it is pursuing artificial general intelligence through deep-learning-guided program synthesis, which it calls Symbolic Learning, and frames the effort as a scientific lab rather than a conventional software company. The public record supports the founder identities and the stated research category through TechCrunch, Y Combinator, Crunchbase, and Ndea; it does not yet show a product, API, pricing, customer, revenue model, or independently reproducible Ndea-authored technical result. The likely buyer, if the company commercializes, could be enterprise research and development, AI-for-science teams, developer-tooling buyers, or model/API customers, but the current buyer and unit of demand remain unknown.

Key Takeaways

Investor reaction
Ndea is exactly the kind of moonshot that deserves attention because François Chollet created Keras and authored ARC-AGI, and Mike Knoop helped scale Zapier; it is also exactly the kind of moonshot where founder prestige can hide missing proof.
Current verdict
hold, pass-leaning, with medium confidence.
Why investors might lean in
The founders are unusually credible for this specific program-synthesis thesis, and adjacent AI value pools are very large, but the public evidence supports founder-market fit more strongly than company progress.
Why investors might pull back
Seventeen-plus months after the January 2025 launch coverage, there is no public product, pricing, reproducible Ndea-authored benchmark, public repository, patent posture, verified funding amount, named research bench, or commercial ownership model.
Highest-leverage fix
Bring one proof pack: a reproducible Ndea-authored benchmark against scaled large-language-model plus test-time search substitutes, a one-page value-capture and proprietary/open intellectual-property boundary, and primary seed economics with allocation terms.
Best next move
Take one time-boxed founder call only if that proof pack is supplied first; convert to pass if the proof pack is not delivered by 2026-09-17.

What Makes This Potentially Fundable

The fundable version of Ndea is not a better chatbot or a general AI services shop. It is a foundation-layer verifiable-reasoning company whose program-synthesis stack can solve generalization tasks that scaled large language models and test-time search cannot, then commercialize that capability through a proprietary platform, license, product, or research factory that customers can pay for and investors can own. That is a legitimate home-run shape because McKinsey sizes broad generative AI economic value in the trillions and Stanford HAI shows AI investment, adoption, and inference-cost curves moving quickly.

The honesty check is that none of those large value pools proves Ndea can capture value. Gartner shows the 2025 generative AI spend forecast is overwhelmingly hardware-heavy, and NVIDIA and Microsoft show current AI economics concentrating upstream in compute and cloud. Ndea must prove both capability and capture, not only ambition.

The Four Holes To Close Before Fundraising

Hole	Investor Fear	What To Bring
No named value-capture mechanism exists.	A genuine research breakthrough could become open science, be replicated by incumbents, or be captured by compute suppliers rather than Ndea shareholders.	A one-page product, license, platform, or spinout-factory model with pricing logic, intellectual-property assignments, and a written proprietary/open boundary.
Seed economics are unresolved.	Crunchbase lists seed investors, while TechCrunch reported at launch that outside capital was not disclosed; without primary terms the check size, ownership, runway, and pro-rata path cannot be modeled.	A primary financing record or signed term summary with amount, valuation, date, use of funds, allocation, and pro-rata rights.
Technical progress is not independently legible.	GitHub shows no public repositories for the company, and public Chollet/ARC materials do not become Ndea company output.	A reproducible Ndea-authored benchmark or demo with protocol and a baseline comparison against o1/AlphaProof-class substitutes.
The team still looks like two famous people, not a lab.	LinkedIn, Y Combinator, and Crunchbase disagree on headcount, and no named senior program-synthesis bench is public.	A payroll-backed org chart, named senior researchers with relevant output, and Knoop's time allocation as CEO.

Decision Snapshot

One-sentence company description: Ndea is a seed-stage frontier AI research lab pursuing artificial general intelligence through deep-learning-guided program synthesis, with exceptional founder-market fit and no public proof yet that the company can produce or own the breakthrough.
Screen: hold, pass-leaning.
Confidence: medium.
Suggested next action: One proof-pack-gated founder call before 2026-09-17.
Why this matters now: If Ndea has a fundable allocation and a real proprietary commercial layer, the entry window could close quickly; if it does not, the public evidence already leans toward pass.
Investor-readiness diagnosis: Founder quality is investor-ready; value capture, technical proof, funding terms, team depth, compute, and governance are not.
Best founder use of this report: Convert the six gating unknowns into documents before any investor conversation.

IC Disagreement Map

The partners reviewed the same Evidence dossier independently: the power-law lens tested asymmetric upside, the prepared-mind lens tested category timing, the founder-jockey lens tested founder and team quality, the risk-reduction lens tested fatal flaws and unsupported claims, and the long-horizon lens tested follow-on logic and compounding durability.

Power-law partner
holdmedium confidence
Ndea has a real fund-returning ceiling if program synthesis becomes a verifiable-reasoning chokepoint, but the research-to-value-capture path and entry economics are not yet underwritable.
Would flip on
A reproducible Ndea-authored result, a named value-capture model, and fund-viable seed economics would move them to pursue; no proprietary commercial layer or no allocation would move them to pass.
Prepared-mind partner
holdmedium confidence
The category inflection may be real because ARC-style generalization remains unsolved, but open-research leakage and substitute progress inside scaled systems keep the structure unresolved.
Would flip on
A Ndea benchmark beating o1/AlphaProof-class substitutes and a clear intellectual-property boundary would move them to pursue; substitute parity would move them to pass.
Founder-jockey partner
holdmedium confidence
The founding pair is unusually credible, but the additional public checks still do not show the senior research bench needed for the hardest build.
Would flip on
Three or more named senior program-synthesis researchers, reconciled headcount, and Knoop full-time would move them to pursue; founders-only execution would move them toward pass.
Risk-reduction partner
passhigh confidence
Value capture, technical progress, and seed economics are all unsupported, and the checked public absences now look adverse rather than neutral.
Would flip on
A value-capture model with intellectual-property assignments, a technical artifact, and verified funding would move them back to hold; open-by-default methods and unreconciled funding would confirm pass.
Long-horizon partner
holdlow confidence
A frontier lab can compound over years, but no compounding asset, milestone framework, funding terms, runway, or compute plan is visible enough to model follow-on logic.
Would flip on
Fund-viable seed terms with pro-rata plus a written 12-18 month milestone plan would move them to pursue; no economics or no milestones would move them to pass.

Round 2 changed the decision texture rather than creating consensus: Priya moved from pursue to hold after public bench evidence remained absent, Lucas stayed pass and increased confidence to high, and the other holds were confirmed unchanged. The final spread is zero pursue, four hold, and one pass, so the disagreement is not social unanimity; it is a live split over whether checked absences are closeable questions or already adverse facts.

Scenario Range

Scenario	What The Company Looks Like In 3-5 Years	Falsifiable Trigger To Watch	Earliest Evidence
Strikeout	Ndea does respected research but no proprietary platform forms, incumbents match the approach through inference-time search, the research bench never scales, and seed holders either own too little or own nothing valuable.	At the next financing, capability is still evidenced only through essays, interviews, and company statements, while there is still no named proprietary layer or primary funding record.	GitHub remains empty, no Ndea-authored benchmark appears, ARC Prize Foundation still lacks a visible conflict firewall, and a frontier incumbent publicly narrows the ARC-style gap.
Base	Ndea becomes a credible research lab with some technical output and a better-articulated commercialization thesis, but commercialization and ownership remain future promises.	A reproducible artifact appears, but it lacks a defended proprietary/open boundary or the round terms still prevent a fund-sized allocation.	A public or private benchmark with protocol, a stated business model, and a primary financing record within 6-18 months.
Home run	Ndea becomes a foundation-layer verifiable-reasoning platform whose symbolic-learning stack leads on generalization and is fenced behind a proprietary license, platform, or product.	A Ndea-authored benchmark beats o1/AlphaProof-class systems and comes with assigned intellectual property, ARC Prize governance separation, and fund-viable pro-rata rights.	A peer-comparable benchmark, patent or assignment documents, written foundation conflict policy, named senior researchers, and priced terms with allocation.

What Investors Will Test

What We Looked For	Current Read	Investor Implication	Founder Prep Priority
A falsifiable home-run mechanism	The mechanism is `unknown`; broad AI value is large, but Ndea has not named what it owns and prices.	Investors will not let AGI ambition substitute for a capture mechanism.	Prepare the product, license, platform, or spinout-factory model and the kill event that disproves it.
Outside public evidence for progress	Founder credentials are `source-backed fact`, but Ndea technical progress remains `unknown`.	Pedigree gets the meeting; artifacts get conviction.	Bring the benchmark, protocol, and baseline comparison.
Team depth beyond the founders	The bench is `unknown` and the headcount is `needs founder validation`.	Investors will underwrite a team, not only two prominent individuals.	Prepare a payroll-backed roster with named senior researchers.
Incentive alignment and capture	The open-research identity is a structural concern rather than a solved asset.	Investors will ask who monetizes a breakthrough: Ndea, incumbents, compute suppliers, or the open field.	Bring intellectual-property assignments and ARC Prize conflict governance.
Double-down and quit triggers	The milestone plan is `unknown`.	A long-horizon lab still needs staged decision gates.	Prepare 12-18 month milestones with symmetric follow-on and stop triggers.

Claim Reconciliation: Inconsistencies Investors Will Catch

Claim Or Metric	Where It Appears	Conflicting / Unreconciled Versions	Why Investors Flag It	How To Reconcile
Seed financing	Crunchbase TechCrunch VentureBeat	Crunchbase lists a seed round with YC, Coatue, and Quiet Capital; January 2025 press says outside capital was not disclosed; no primary public amount is available.	Funding timing, valuation, and ownership determine whether a small fund can participate at all.	Provide dated closing documents, primary filing, or a signed financing summary.
Seed amount	Crunchbase	Crunchbase public row did not provide a round amount; snippet-grade paid-database leads are not reliable enough to cite as fact.	A large round could make the ownership path impossible; an unknown round makes runway impossible to model.	Provide the amount, valuation, date, use of funds, allocation, and pro-rata rights.
Team size	Y Combinator LinkedIn Crunchbase	YC lists team size 15, LinkedIn shows 14 associated employees, and Crunchbase shows a smaller employee range.	The company is being underwritten as a research lab; true bench depth is central.	Provide a payroll-backed org chart by role, location, and full-time status.
Founder/operator commitment	Mike Knoop TechCrunch Zapier	Knoop titles himself CEO of Ndea while also listing board roles at Zapier and ARC Prize Foundation.	The operator half of the founding pair is valuable only if sufficiently hands-on.	Provide time allocation, responsibilities, and board/governance commitments.
Benchmark governance	Ndea About ARC Prize Foundation	The founders are associated with both the company and the benchmark ecosystem; no public conflict policy or firewall was found.	Ndea results on ARC-style benchmarks will be questioned without independence.	Provide bylaws, independent governance, and conflict-disclosure policy.

Diligence Findings: Issues To Fix Before You Raise

1 deal-breaker4 high2 medium

Ndea has no falsifiable value-capture mechanism in the public record.
deal-breaker
Commercial and legal diligence
Fix
Name the product, license, platform, or factory mechanism, show pricing logic, and document proprietary ownership.
Public technical progress is not independently legible.
high
Technical diligence
Fix
Prepare a reproducible Ndea-authored result with protocol, baselines, and ownership trail.
Seed economics cannot be reconciled from public sources.
high
Financial diligence
Fix
Provide primary terms, amount, date, valuation, use of funds, allocation, and pro-rata.
Founder-led open benchmark governance creates a conflict and value-leakage concern.
high
Legal and governance diligence
Fix
Provide ARC Prize governance separation, bylaws, and conflict policy.
Team depth beyond the two founders is not public.
high
Team diligence
Fix
Show named senior researchers, employment status, and role ownership.
Compute supply and runway are not evidenced.
medium
Technical and financial diligence
Fix
Provide vendor agreement, credits, committed capacity, burn forecast, and runway.
Public legal, privacy, security, and terms surfaces were not found.
medium
Legal and procurement diligence
Fix
Prepare basic corporate, data, security, and intellectual-property materials even if the company is pre-product.

Data Room Readiness Checklist

Procurement-readiness score: not ready.
Rationale: Ndea appears pre-product and research-stage, but a frontier AI buyer, partner, or enterprise research customer would still expect legal, intellectual-property, security, model-governance, data-use, and compute-risk materials that are not visible publicly.

Data Room Area	What Investors Expect	Current Read	Status	Priority
Company overview & corporate	One-pager, incorporation, good standing, structure	Public brand, mission, founders, and accelerator profile exist, but legal entity details are not confirmed.	`partial`	High
Financials	Monthly P&L (18-24 mo), 3-yr model with assumptions, burn/runway	No public revenue, burn, runway, or funding amount is available.	`missing`	Critical
Cap table & funding history	Clean cap table, prior rounds, SAFEs/notes, 409A	Crunchbase lists seed investors, but the terms and amount are `unknown`.	`partial`	Critical
Legal & IP	Bylaws, board consents, IP assignments (founders + contractors), material contracts	No public patent posture, assignment pack, or ARC Prize firewall is visible.	`missing`	Critical
Product & technology	Roadmap, architecture overview, security/compliance docs	The thesis is public, but there is no reproducible Ndea-authored artifact or architecture proof.	`partial`	Critical
Team	Org chart, key employment/advisor agreements, vesting	Founders are supported; bench and time allocation are unresolved.	`partial`	High
Customers & traction	Retention cohorts, ARR/MRR bridge, sales funnel, 3-5 references	No customers, pilots, revenue, or product traction are public.	`missing`	Medium
Market & competition	TAM/SAM/SOM, competitive landscape, pricing	Adjacent market proxies exist, but Ndea-specific SAM/SOM and pricing are unavailable.	`partial`	High

First-Call Agenda For The Startup

Time	Topic	Founder Goal	Evidence To Bring
0-5 minutes	State the investable hypothesis.	Define the one fund-returning path in plain terms.	One-page value-capture model and explicit kill event.
5-15 minutes	Technical proof and substitute comparison.	Show that Ndea's approach is distinct from scaled large-language-model plus test-time search.	Reproducible benchmark, protocol, baseline results, and ownership trail.
15-25 minutes	Funding terms and allocation.	Prove the round is investable for a small fund.	Primary financing summary, valuation, allocation, and pro-rata rights.
25-35 minutes	Team and compute.	Show the lab can execute beyond founder reputation.	Org chart, named senior researchers, Knoop time allocation, compute contract, and runway model.
35-45 minutes	Governance and next gate.	Resolve ARC Prize conflict and define follow-on milestones.	Foundation conflict policy, intellectual-property assignments, and 12-18 month milestone plan.

Business Model And Revenue Signals

Company says it will invent and commercialize ideas.
Company-controlled pagehigh
SourceNdea
This states ambition, not a revenue model.
Contact page includes services language.
Company-controlled pagemedium
SourceNdea Contact
It is not a priced services offering.
Seed investors are listed publicly.
source-backed fact for investors; unknown for amountPublic databasemedium
SourceCrunchbase
Financing quality is a signal; economics are not.

Ndea and Crunchbase were checked and showed no public product, pricing, docs, API, revenue, customers, pilots, or grants. These absences carry moderate weight for a pre-product research lab but still require founder confirmation of pre-commercial status.

Revenue or profit is not publicly available. Any revenue scenario would require assumptions about product, buyer, pricing, and margin that are not supported by public evidence.

Revenue Quality Checklist

Before this can move from hold to pursue, Ndea should be ready to show whether revenue exists, whether any contracts are paid or pilot-only, what the first commercial unit will be, gross margin after compute, customer concentration, renewal logic, and how customer contracts connect to intellectual-property ownership. Investors will rebuild customer acquisition cost, lifetime value, payback, gross revenue retention, net revenue retention, and churn only if a commercial model exists; until then the analogous proof is research milestone cadence, compute burn, runway, and follow-on requirements.

Funding And Ownership Context

Item	Public Read	Evidence Label	Diligence Request
Total raised / rounds	Crunchbase lists a seed round; no primary public amount was verified.	`source-backed fact` for seed row; `unknown` for amount	Provide primary financing record with amount and date.
Instruments (equity / SAFE / notes)	Not public.	`unknown`	Provide term summary and instrument details.
Investors	Crunchbase lists Y Combinator, Coatue, and Quiet Capital.	`source-backed fact`	Confirm allocation, lead, and pro-rata rights.
Post-money valuation	Not public.	`unknown`	Provide valuation and ownership math.
Cap table / ownership	Not public.	`unknown`	Provide cap table, founder ownership, option pool, SAFEs or notes, and pro-rata path.
Burn & runway	Not public.	`unknown`	Provide monthly burn, compute spend, committed credits, and runway.
Next-round plan	Not public.	`unknown`	Provide milestone plan and follow-on financing target.

SEC EDGAR and Google Search were checked and showed no public Form D filing for Ndea. This absence carries little weight because seed rounds often stay private until a later disclosure.

Founder And Team

François Chollet
Co-founder; public technical lead for Keras and ARC-AGI-related work before Ndea.
LinkedIn GitHub GitHub Repositories TechCrunch
Evidencesource-backed fact for identity and public technical footprint
Confidencehigh
Mike Knoop
Co-founder; self-titles as CEO; previously co-founded Zapier.
Mike Knoop TechCrunch Zapier Y Combinator
Evidencesource-backed fact for founder identity and self-stated title; needs founder validation for time commitment
Confidencehigh
Team size
Public directories disagree on Ndea's headcount.
Y Combinator LinkedIn Crunchbase
Evidenceneeds founder validation
Confidencelow
Hiring signal
Ndea lists a Search Guidance role and a referral program.
Ndea Jobs Ndea Search Guidance role Ndea Referral Program
Confidencehigh

Founder Competency Coverage

Domain depth
François Chollet
evidenced
Mike Knoop
claimed
What The Evidence Is
Chollet's public repositories and launch press support deep learning and ARC-AGI domain depth; Knoop's program-synthesis depth is not public.
Technical build capability
François Chollet
evidenced
Mike Knoop
claimed
What The Evidence Is
Chollet has public technical artifacts; Knoop's technical capability is described on company-controlled pages and press, but no technical artifact was found.
Product
François Chollet
claimed
Mike Knoop
evidenced
What The Evidence Is
Knoop's Zapier history supports product/operator experience; Chollet's Keras product impact is visible but not a commercial company outcome.
GTM / sales
François Chollet
absent
Mike Knoop
evidenced
What The Evidence Is
Zapier scale supports Knoop's go-to-market exposure; Chollet has no public Ndea go-to-market record.
Leadership / hiring
François Chollet
claimed
Mike Knoop
evidenced
What The Evidence Is
Knoop's Zapier operating background supports leadership; Ndea's actual team build remains unverified.
Fundraising history
François Chollet
claimed
Mike Knoop
claimed
What The Evidence Is
Crunchbase lists seed investors, but the funding amount and terms are not primary-verified.
Prior founding outcomes
François Chollet
evidenced
Mike Knoop
evidenced
What The Evidence Is
Chollet's Keras and ARC-AGI public artifacts and Knoop's Zapier co-founder history are supported by public sources.

Deep-learning-guided program synthesis
High
Team coverage today
Chollet coverage is evidenced; bench beyond him is unknown.
What would close it
Named researchers, artifacts, and division of research ownership.
Search-guidance and symbolic-reasoning stack
High
Team coverage today
Public role suggests hiring intent; filled ownership is unknown.
What would close it
Named owners with relevant publications or shipped systems.
Commercialization factory
High
Team coverage today
Knoop's operator background is evidenced; Ndea buyer and model are unknown.
What would close it
First product or license plan with buyer, pricing, and sales owner.
Compute-intensive frontier execution
Medium
Team coverage today
Public claim only.
What would close it
Contracted compute supply and burn/runway model.

Public Professional Footprint

These checks carry low decision weight by design: they corroborate or weaken the founder-coverage table and sharpen founder-call questions, but none alone changes the verdict.

Technical artifacts (repo level)
François Chollet
evidenced
Mike Knoop
absent
What The Public Record Shows
GitHub Repositories show public technical work for Chollet; no Knoop technical repository was found.
Professional social content
François Chollet
evidenced
Mike Knoop
claimed
What The Public Record Shows
Chollet's LinkedIn dataset shows research-oriented activity; Knoop's public X handle appears via Y Combinator, but posts were not retrieved from public sources.
Education / credentials
François Chollet
unknown
Mike Knoop
unknown
What The Public Record Shows
The public evidence used here did not independently verify degrees or credentials.
Publications / patents / certifications
François Chollet
evidenced
Mike Knoop
absent
What The Public Record Shows
Chollet's public ARC/Keras materials are visible; no Ndea patent posture was found in public database checks, and Knoop's publications were not evidenced.

Net read: founder-market fit is the reason to take the call, while the public footprint still makes the call a proof-pack gate rather than a conviction meeting.

Founder-Market Fit Read

What is promising: Chollet has unusually direct technical relevance to ARC-style generalization, and Knoop brings rare scaled-operator credibility from Zapier.
What is missing: The public record does not yet show Ndea's own technical output, senior bench, Knoop's time allocation, or a commercialization owner for the first wedge.
What to prepare: A team roster, role ownership map, founder time commitments, technical artifact, and commercialization plan.

ARC Prize Foundation
Related nonprofit benchmark ecosystem with founder overlap.
Ndea About ARC Prize Foundation
Evidencesource-backed fact for public roles; unknown for governance separation
Confidencemedium
Follow-upProvide bylaws, independent governance, and conflict-disclosure policy.
Zapier
Knoop co-founded Zapier and remains publicly associated with it.
TechCrunch Zapier Mike Knoop
Evidencesource-backed fact for prior-venture association; needs founder validation for time split
Confidencehigh
Follow-upConfirm Knoop's time allocation and operating responsibilities at Ndea.
Keras / ARC-AGI repositories
Chollet's pre-Ndea technical artifacts.
GitHub
Evidencesource-backed fact
Confidencehigh
Follow-upShow which methods, if any, are assigned to Ndea.

Traction

Customer Status Table

Customer	Strongest Evidence	Customer-Controlled?	Status	Paid?	Seats	Renewal	Confidence

Ndea, Crunchbase, TechCrunch, and VentureBeat were checked and showed no public customer logo, case study, paid pilot, grant award, or deployment. These absences carry moderate weight for a research lab but still require founder confirmation of pre-commercial status.

Traction Signals

Seed investors listed as Y Combinator, Coatue, and Quiet Capital.
source-backed fact for investor names; unknown for amountmedium
SourceCrunchbase
Primary financing terms and allocation.
Y Combinator Winter 2026 profile exists.
source-backed facthigh
SourceY Combinator
Confirm batch participation details and team size.
LinkedIn company page shows 1,848 followers and 14 associated employees.
source-backed factmedium
SourceLinkedIn
Payroll-backed headcount.
YouTube channel shows 1,080 subscribers, 53 videos, and 48,722 views at retrieval.
source-backed factmedium
SourceYouTube
Treat as content cadence only, not traction.
Crunchbase includes a web-traffic proxy of 4,772 monthly visits.
source-backed factmedium
SourceCrunchbase
Treat as a directional aggregator metric only.

Hiring And Org Momentum

Company careers page
One visible technical role and general hiring copy.
SourceNdea Jobs
Active hiring intent, not evidence of a scaled bench.
Company role page
Technical Staff, Search Guidance role references an aggressive compute budget.
SourceNdea Search Guidance role
Role mix aligns with thesis, but filled capacity remains unknown.
Referral program
$10,000 referral program was visible.
SourceNdea Referral Program
Recruiting effort is active; senior bench remains unevidenced.

LinkedIn Jobs, Indeed, and Glassdoor were checked and showed no Ndea listings. This adds little weight to the hiring read because the company careers page already lists an open role and job boards often lag direct postings, while employee sentiment remains unavailable.

Role-mix read: The visible hiring is technical and search-guidance oriented, which fits the thesis, but there is no public evidence that senior program-synthesis roles are filled.
Momentum read (growth / steady / contraction / unknown): unknown; the public record shows hiring intent, not actual hiring velocity.

Traction Quality Read

Traction Dimension	Current Status	Good Enough For First Call?	Needed For Deep Diligence
Funding	Seed investors listed, amount unknown.	Yes, only as a gating topic.	Primary funding terms and allocation.
Product	No public product, pricing, docs, or API.	Yes for a research lab, but only if technical artifact exists.	Product roadmap or research-to-product path.
Customers	No public customers or pilots found.	Not necessary for a first research-lab call.	If commercialization is claimed, provide contracts, pilots, grants, or references.
Technical progress	No public Ndea-authored artifact found.	Not enough without a private proof pack.	Reproducible benchmark or demo.
Team	Founders strong; bench unknown.	Yes, if team roster is supplied.	Named senior researchers and employment evidence.

Competitive Landscape

Segment	Examples	Customer Alternative	Pressure On Company
Frontier AI incumbents	OpenAI, Anthropic, Google DeepMind, xAI	Use foundation-model APIs, reasoning systems, or bundled enterprise AI platforms.	They can spend more, recruit more, and absorb program-synthesis ideas into existing stacks.
Independent AGI labs	Safe Superintelligence and Ndea	Join or fund another research-first AGI lab.	Talent and capital compete on founder credibility before products exist.
Benchmark and open-source ecosystem	ARC Prize Foundation and ARC entrants	Build against open benchmarks and open methods instead of buying from Ndea.	The ecosystem can validate the thesis while commoditizing methods.
Academic and corporate neurosymbolic research	Microsoft Research, IBM, university groups, and arXiv-visible researchers	Wait for papers, open-source implementations, or incumbent research translation.	Public research can reduce Ndea's proprietary edge.
Compute and cloud platforms	NVIDIA, Microsoft Azure, Google Cloud, and other hyperscalers	Pay for compute and frontier platform tooling rather than a small lab's software.	Upstream suppliers may capture much of the economic value.

Category Visibility Snapshot

Google / USprogram synthesis AGI AI research lab

Captured page-one results (2026-06-17): Hugging Face, ARC Prize, Reddit, TechCrunch, arXiv, and related research pages occupied the visible results.
Was the company present?: Ndea was not a top organic domain result; it appeared indirectly through press context.

Google / UKprogram synthesis AGI AI research lab

Captured page-one results (2026-06-17): Hugging Face, ARC Prize, Reddit, TechCrunch, arXiv, and related research pages showed a similar pattern.
Was the company present?: Ndea was not a top organic domain result.

Google / USneurosymbolic AI program synthesis AGI

Captured page-one results (2026-06-17): Reddit, ARC Prize, Medium/Turing Post, arXiv, neurosymbolic.org, and IBM appeared.
Was the company present?: Ndea appeared only as a secondary mention in category commentary.

Google / USARC-AGI benchmark AI labs

Captured page-one results (2026-06-17): ARC Prize, arXiv, Kaggle, and press dominated the result set.
Was the company present?: Ndea.com did not own the query.

Google / USNdea AI research lab Chollet

Captured page-one results (2026-06-17): Ndea, Y Combinator, TechCrunch, and VentureBeat appeared prominently.
Was the company present?: Yes, Ndea owned the branded query.

Google Search (US and UK) was checked and Ndea did not appear on page one for the generic category queries reviewed. This carries low weight for market share but matters for category ownership and recruiting visibility.

Visibility read (inference, low confidence): Ndea has strong branded visibility but weak generic category visibility; this matters for recruiting, partner discovery, and perceived category leadership, not for proof of technology.

Pricing And Competitive Benchmark

Alternative	What It Offers	Public Price Signal	Price Vs. This Company	Evidence Label
Ndea	Research lab pursuing program synthesis for AGI.	No public price, product, API, or services rate was found.	Ndea's price is `unknown`.	`unknown` via Ndea
OpenAI	Foundation models, ChatGPT, and developer-facing AI platform.	Public homepage confirms commercial product surface; this report did not rely on a pricing page.	Ndea cannot be benchmarked until it has a product or contract unit.	`company claim` via OpenAI
Anthropic	Claude and frontier model products.	Public homepage confirms commercial product surface; this report did not rely on a pricing page.	Ndea cannot be benchmarked until it has a product or contract unit.	`company claim` via Anthropic
Google DeepMind	Gemini and scientific AI within Google's ecosystem.	Public homepage confirms product and research surface; this report did not rely on a pricing page.	Ndea competes for research attention more than price today.	`company claim` via Google DeepMind
xAI	Grok models, API, and developer console.	Public homepage confirms commercial model/API surface; this report did not rely on a pricing page.	Ndea has no comparable public package.	`company claim` via xAI
ARC Prize Foundation	Open benchmark and prize ecosystem.	Free/open benchmark ecosystem and prize structure.	It is not a paid substitute; it competes through open research and mindshare.	`company claim` via ARC Prize Foundation

Price positioning read: Ndea is not price-positioned because it has no public commercial unit.
Price claims to correct or substantiate: Any investor-facing claim about pricing, services, licensing, or customer willingness to pay should be supported with a rate card, contract, or clear roadmap.

Competitive Wedge

Ndea's possible wedge is founder-native insight into ARC-style generalization and a focused program-synthesis architecture that incumbents treat as only one component. The wedge is currently copyable or absorbable because the public record does not show a protected implementation, a benchmark lead, an intellectual-property boundary, or a proprietary dataset/method asset. The first defensible wedge would be a Ndea-authored result that scaled large-language-model search does not match, paired with proprietary ownership of the method.

Risks And Open Questions

Risk	Severity	Evidence	What To Ask
Value capture leaks to open science or incumbents.	Critical	ARC Prize Foundation public mission and founder overlap; no public product or pricing on Ndea.	What exactly stays proprietary and how is it monetized?
Substitute path absorbs the thesis.	Critical	ARC Prize Foundation discusses o1 and AlphaProof-style progress on ARC.	What benchmark shows Ndea's dedicated stack leads?
Entry economics fail fund-return math.	Critical	Crunchbase lists investors but not amount; January 2025 press did not disclose capital.	What allocation and pro-rata exist at fund-check size?
Team is founders-only on the core build.	High	GitHub has no public members or repositories; headcount signals conflict.	Who are the named senior researchers and what have they shipped?
Compute runway is not evidenced.	High	Ndea Search Guidance role references compute, but no vendor or capacity is public.	What committed compute and runway does the seed buy?
Governance conflict affects benchmark credibility.	Medium	Ndea About and ARC Prize Foundation show founder overlap.	What written firewall governs ARC Prize and Ndea?

Pre-Mortem: The Most Likely Obituary

Cause of death (one sentence): Ndea did respected program-synthesis research but failed as a venture-backed company because incumbents matched the approach through test-time search, the best methods flowed into open science, and the lab never formed a proprietary, priced layer before its capital base ran out.
The causal chain (3-5 steps from today to the shutdown): First, the lab stays pre-product while compute burn grows. Second, OpenAI, DeepMind, and other incumbents keep closing the ARC-style gap with scaled systems and search. Third, Ndea publishes or shares the novel parts through the founders' open-research gravity rather than fencing them as company intellectual property. Fourth, the senior bench and buyer wedge do not materialize quickly enough for follow-on investors. Fifth, the company becomes an acqui-hire or a respected research footnote rather than a fund-returning platform.
The earliest observable warning sign: The next financing still has no reproducible Ndea-authored benchmark and no named proprietary commercial layer.
The question that defuses this chain today: What can Ndea own and price that the open field and frontier incumbents cannot replicate, and what proof shows it works?

Decision-Critical Unknowns

Unknown	Why It Is Decision-Critical	Best Evidence	Decision Effect
Value-capture mechanism	It determines whether a breakthrough can return capital.	Product/license/platform plan, pricing, intellectual-property assignments, and ARC Prize firewall.	A defended mechanism moves toward pursue; no mechanism moves to pass.
Seed terms and allocation	It determines whether the fund can own enough and defend pro-rata.	Primary financing record and allocation rights.	Fund-viable terms move toward pursue; no allocation or unreconciled funding moves to pass.
Technical artifact	It determines whether Ndea has progress beyond founder narrative.	Reproducible benchmark or demo with protocol and baselines.	A strong artifact moves toward pursue; no artifact sustains pass-leaning hold.
Senior research bench	It determines whether the company can execute beyond two founders.	Org chart and named senior researchers with relevant output.	Confirmed bench moves founder-jockey view up; founders-only bench moves down.
Compute and runway	It determines feasibility for a capital-intensive lab.	Signed compute agreement and burn/runway model.	Contracted capacity supports diligence; no runway math keeps capital-intensity kill live.
Governance separation	It determines benchmark credibility and open-research leakage risk.	Foundation bylaws, independent board, and conflict policy.	Clear firewall reduces risk; no firewall sustains the value-capture concern.

Diligence Questions

First Call

Question	Why It Matters	Good Evidence
What exactly does Ndea own and sell if the research works?	This is the primary value-capture gate.	One-page capture model, pricing logic, intellectual-property assignments, and proprietary/open boundary.
What reproducible Ndea-authored result exists since launch?	Capability is currently `company claim` rather than demonstrated company output.	Benchmark or demo with protocol, baselines, and repeatability.
What are the seed amount, valuation, date, and allocation rights?	The fund cannot model ownership, runway, or follow-on without them.	Primary financing document or signed summary.
Who are the senior researchers beyond Chollet and Knoop?	The hardest build requires a bench.	Named roster, role ownership, and relevant output.
How is ARC Prize governance separated from Ndea's commercial interests?	Benchmark credibility and value capture require a firewall.	Bylaws, independent directors, and conflict-disclosure policy.

Follow-Up

Question	Why It Matters	Good Evidence
What compute capacity is contracted, and what runway does the seed buy?	The technical path is compute-intensive.	Signed cloud/GPU terms and burn/runway model.
What is Knoop's time allocation as CEO?	The operator pairing matters only if hands-on.	Employment terms and operating responsibilities.
What buyer and first wedge would pay for the first output?	The company needs a route from research to revenue.	Named segment, willingness-to-pay proof, and early sales funnel.
What 12-18 month milestone justifies a follow-on?	A long-horizon lab needs disciplined staging.	Written milestones with double-down and quit triggers.
How will Ndea handle data, security, and model-governance obligations if it commercializes?	Enterprise or regulated buyers will demand procurement readiness.	Security, privacy, data retention, model-risk, and compliance materials.

Kill Criteria

Kill Criterion	Evidence That Would Trigger It
Value-capture kill	Methods default open with no proprietary boundary, no product/license layer, and no ARC Prize firewall.
Substitute-path kill	o1/AlphaProof-class systems match or exceed Ndea's claimed approach without a dedicated symbolic-learning stack.
Entry-economics kill	The round is too large or closed for fund-check allocation, or there is no pro-rata path.
Funding-fact reconciliation kill	The public funding conflict cannot be reconciled with a primary record.
Execution-velocity kill	There is no reproducible Ndea-authored artifact after the proof-pack gate.
Two-person-shop kill	The bench is essentially founders-only and Knoop is not full-time operationally engaged.
Capital-intensity kill	No committed compute capacity or runway math exists for the technical plan.

Double-Down Criteria

Double-Down Criterion	Evidence That Would Justify More Diligence
Technical lead	Reproducible Ndea-authored benchmark beating scaled-model search on ARC or adjacent generalization tasks.
Ownable capture	Product/license/platform plan with intellectual-property assignments and closed/open boundary.
Investable round	Allocation at $500K-$2.5M with pro-rata and enough ownership path to matter.
Scalable team	Three or more named senior researchers with relevant output and Knoop confirmed full-time.
Feasible runway	Signed compute agreement and burn/runway plan tied to milestones.
Governance clarity	ARC Prize Foundation firewall and conflict policy.

Founder Action Plan

Timeframe	Action	Output
Before next investor call	Reconcile seed economics and allocation.	Primary financing record, cap-table summary, valuation, use of funds, allocation, and pro-rata rights.
Before next investor call	Articulate the proprietary value-capture model.	One-page capture model, pricing logic, intellectual-property assignments, and open/closed boundary.
Before next investor call	Prepare the technical proof pack.	Reproducible benchmark or demo with protocol and baselines against scaled large-language-model search.
Near term	Prove team depth and founder commitment.	Named senior researcher roster, role ownership, payroll-backed headcount, and Knoop time allocation.
Near term	Confirm compute feasibility.	Signed compute or cloud agreement, committed capacity, burn forecast, and runway months.
12-18 month horizon	Define staged option gates.	Written double-down and quit milestones, plus ARC Prize governance separation.

Decision

Screen: hold, pass-leaning.
Confidence: medium.
Rationale: Ndea is first-call-worthy because the founding pair is unusually credible for the exact technical thesis, but it is not deep-diligence-ready because public evidence reinforces gaps in value capture, technical proof, seed economics, team depth, compute, and governance.
What would move this to pursue: A reproducible Ndea-authored result, a named and assigned proprietary commercial layer with ARC Prize firewall, fund-viable seed terms with allocation and pro-rata, named senior researchers, and a compute-backed 12-18 month milestone plan.
What would move this to pass: No named proprietary commercial layer, no funding reconciliation, no fund-check allocation, no technical artifact, founders-only bench, or substitute systems matching the approach through inference-time search.
Recommended next step: One proof-pack-gated founder call before 2026-09-17.
Founder preparation standard: Do not take a general vision call; bring documents that collapse the six gating unknowns.

How We Would Miss This One

The miss scenario: Ndea could be the one that got away if the public record is simply too early and the founders already have a private benchmark, a proprietary method boundary, a real senior bench, and allocation in a round that will never again be accessible at seed economics.
Flip conditions: Technical proof, value-capture proof, verified economics with allocation, bench proof, compute proof, and governance proof move this from pass-leaning hold toward pursue.
Revisit trigger / date: Revisit immediately on receipt of the proof pack or a primary funding record; otherwise convert the watch to pass after 2026-09-17.

Source Log

Ndea homepage
ndea.com
Company thesis, mission, category framing, factory language, and social links.
Retrieved 2026-06-17Company-controlled pagehigh
Ndea About
ndea.com
Founder names, company-stated bios, and ARC Prize relationship.
Retrieved 2026-06-17Company-controlled pagehigh
Ndea Jobs
ndea.com
Hiring intent, symbolic-learning language, and role geography.
Retrieved 2026-06-17Company-controlled careers pagehigh
Ndea Search Guidance role
ndea.com
Open technical role and compute-budget claim.
Retrieved 2026-06-17Company-controlled job posthigh
Ndea Referral Program
ndea.com
Recruiting referral signal.
Retrieved 2026-06-17Company-controlled pagehigh
Ndea Contact
ndea.com
Contact and services language.
Retrieved 2026-06-17Company-controlled pagehigh
TechCrunch Ndea launch article
techcrunch.com
Founder identities, Knoop/Zapier link, launch timing, funding non-disclosure, and program-synthesis context.
Retrieved 2026-06-17Reputable presshigh
VentureBeat Ndea profile
venturebeat.com
Launch-time funding non-disclosure and company thesis context.
Retrieved 2026-06-17Reputable presshigh
Y Combinator Ndea profile
ycombinator.com
Accelerator profile, founder roster, team size, batch, and public profile links.
Retrieved 2026-06-17Accelerator directoryhigh
Crunchbase Ndea profile
crunchbase.com
Seed investor row, employee range, web-traffic proxy, patent count field, and company categorization.
Retrieved 2026-06-17Startup databasemedium
LinkedIn Ndea company page
linkedin.com
Company headcount proxy, follower count, HQ, and hiring signal.
Retrieved 2026-06-17Platform profile datasetmedium
LinkedIn François Chollet profile
linkedin.com
Chollet's Ndea current-company link and professional activity signal.
Retrieved 2026-06-17Public professional profilehigh
François Chollet GitHub profile
github.com
Technical footprint and public developer identity.
Retrieved 2026-06-17Technical artifacthigh
François Chollet GitHub repositories
github.com
Keras, ARC-AGI, and related public repository footprint.
Retrieved 2026-06-17Technical artifacthigh
Ndea GitHub organization
github.com
Absence of public Ndea repositories and visible members.
Retrieved 2026-06-17Technical artifact absencehigh
Mike Knoop personal site
mikeknoop.com
Knoop's self-stated Ndea CEO title and concurrent board roles.
Retrieved 2026-06-17Founder-controlled public pagemedium
Zapier About
zapier.com
Zapier scale as context for Knoop's prior venture.
Retrieved 2026-06-17Prior-employer pagehigh
ARC Prize Foundation About
arcprize.org
Open-source benchmark ecosystem mission, founder overlap, and governance-risk context.
Retrieved 2026-06-17Related nonprofit pagemedium
ARC Prize program-synthesis blog
arcprize.org
Substitute-path risk, o1/AlphaProof context, and program-synthesis technical framing.
Retrieved 2026-06-17Related technical blogmedium
YouTube Ndea channel
youtube.com
Public content cadence and audience proxy.
Retrieved 2026-06-17Platform datasetmedium
Gartner GenAI spending forecast
gartner.com
Generative AI spend envelope and hardware/software/services split.
Retrieved 2026-06-17Analyst report press releasehigh
McKinsey economic potential of generative AI
mckinsey.com
Economic-value pool, function split, and research and development value context.
Retrieved 2026-06-17Analyst reporthigh
McKinsey gen AI in life sciences
mckinsey.com
Life-sciences generative AI value proxy.
Retrieved 2026-06-17Analyst reportmedium
Stanford HAI AI Index 2025
hai-production.s3.amazonaws.com
AI investment, adoption, inference-cost curve, and industry-led model supply.
Retrieved 2026-06-17Academic and institutional reporthigh
NVIDIA 2025 Annual Report
s201.q4cdn.com
Compute supplier revenue and gross margin benchmark.
Retrieved 2026-06-17Public-company filinghigh
Microsoft Annual Report 2025
microsoft.com
Cloud revenue, gross margin, capex, and AI platform adoption benchmark.
Retrieved 2026-06-17Public-company filinghigh
Alphabet 2024 Form 10-K
sec.gov
Capex, research and development, and Google Cloud benchmark.
Retrieved 2026-06-17Public-company filinghigh
Meta Platforms 2024 Form 10-K
sec.gov
Long-horizon moonshot investment benchmark and capex/research and development context.
Retrieved 2026-06-17Public-company filinghigh
EC General-Purpose AI Models in the AI Act Q&A
digital-strategy.ec.europa.eu
General-purpose AI obligations, systemic-risk thresholds, and future regulatory exposure.
Retrieved 2026-06-17Official regulator Q&Ahigh
EU AI Act Article 55 mirror
artificialintelligenceact.eu
Systemic-risk duties context.
Retrieved 2026-06-17Regulatory text mirrorhigh
California SB 53 bill text
leginfo.legislature.ca.gov
Frontier-model threshold and catastrophic-risk definitions.
Retrieved 2026-06-17Official legislative texthigh
BIS export-control press release
bis.gov
Advanced-computing chip, high-bandwidth-memory, and tool-control context.
Retrieved 2026-06-17Official regulator sourcehigh
UT Austin neurosymbolic program-synthesis survey
cs.utexas.edu
Program-synthesis and neurosymbolic AI definition and technique context.
Retrieved 2026-06-17Academic paperhigh
Nature AI for Science 2025
nature.com
AI-for-science category definition.
Retrieved 2026-06-17Academic and institutional sourcehigh
Royal Society Science in the Age of AI
royalsociety.org
Scientific AI adoption barriers and context.
Retrieved 2026-06-17Institutional reportmedium
OpenAI homepage
openai.com
Frontier incumbent positioning.
Retrieved 2026-06-17Competitor company pagehigh
Anthropic homepage
anthropic.com
Frontier incumbent positioning.
Retrieved 2026-06-17Competitor company pagehigh
Google DeepMind homepage
deepmind.google
Frontier and scientific AI incumbent positioning.
Retrieved 2026-06-17Competitor company pagehigh
xAI homepage
x.ai
Frontier model, API, and developer-console positioning.
Retrieved 2026-06-17Competitor company pagehigh
Safe Superintelligence homepage
ssi.inc
Independent AGI lab peer positioning.
Retrieved 2026-06-17Competitor company pagehigh
Crunchbase OpenAI profile
crunchbase.com
Employee and investor scale proxy for an incumbent lab.
Retrieved 2026-06-17Startup databasemedium
Crunchbase Anthropic profile
crunchbase.com
Employee and investor scale proxy for an incumbent lab.
Retrieved 2026-06-17Startup databasemedium
Crunchbase xAI profile
crunchbase.com
Employee and investor scale proxy and corporate-structure context.
Retrieved 2026-06-17Startup databasemedium

Access limitations: Mike Knoop's LinkedIn profile could not be read from public sources, so his employment timeline and full-time commitment remain needs founder validation. Ndea's X account, several search-result-only funding leads, and paid database snippets were excluded from cited evidence, so social reach and the unverified seed amount are not treated as facts. No public product, pricing, docs, privacy, terms, security, legal, customer, Form D, compute-vendor, or ARC Prize firewall page was available in the public record, which limits this report to a founder-quality and proof-pack readiness judgment rather than a complete diligence conclusion.

Public disclaimer: This is an independent public teardown based solely on publicly available information retrieved on 2026-06-17. It is research support, not investment, legal, tax, or financial advice. Company-provided statements are labeled as claims, not verified facts. Ndea did not participate in or review this report. Errors and omissions are possible; final decisions remain with the reader.

What The Company Does#

Key Takeaways#

What Makes This Potentially Fundable#

The Four Holes To Close Before Fundraising#

Decision Snapshot#

IC Disagreement Map#

Scenario Range#

What Investors Will Test#

Claim Reconciliation: Inconsistencies Investors Will Catch#

Diligence Findings: Issues To Fix Before You Raise#

Data Room Readiness Checklist#

First-Call Agenda For The Startup#

Business Model And Revenue Signals#

Revenue Quality Checklist#

Funding And Ownership Context#

Founder, Team, And Related Entities#

Founder And Team#

Founder Competency Coverage#

Public Professional Footprint#

Founder-Market Fit Read#

Related Entities And Founder-Associated Companies#

Traction#

Customer Status Table#

Traction Signals#

Hiring And Org Momentum#

Traction Quality Read#

Competitive Landscape#

Category Visibility Snapshot#

Pricing And Competitive Benchmark#

Competitive Wedge#

Risks And Open Questions#

Pre-Mortem: The Most Likely Obituary#

Decision-Critical Unknowns#

Diligence Questions#

First Call#

Follow-Up#

Kill Criteria#

Double-Down Criteria#

Founder Action Plan#

Decision#

How We Would Miss This One#

Source Log#