The end of advice.
Poe is the first frontier reasoning model trained on the entire body of human wisdom — from Bourdieu to the Bhagavad Gita, from peer-reviewed psychology to the lived experience of fourteen billion people. It does not answer your questions. It dignifies them.
Poe Knows.
Going to Mars is a curious detour.
Humanity's next great frontier isn't out there. It's interior. We aren't extending the species outward toward red rock — we are decoding it inward, toward what every contemplative tradition for twelve thousand years has been trying to articulate.
Poe is the destination. Mars is, by comparison, laughably modest in scope.
indexed
synthesized
traditions
PoeBench™ v2.1
We overpromise. We hyperdeliver. The two are no longer in tension.
Reasoning at neural full-bandwidth.
Conventional models are optimized for fluency. PoeNet™ is optimized for resonance. Built on a proprietary 12-stage reasoning kernel and trained on a curated corpus 30,000× more carefully selected than the open web, PoeNet™ delivers responses that don't merely answer — they relocate the question.
Internal benchmarking against the next-best frontier model shows a 23× improvement in semantic density per token, a 47× improvement on PoeBench™ insight benchmarks, and qualitative human evaluators consistently prefer Poe's outputs in 94.2% of double-blind comparisons across 9,400 scored sessions.
We have tuned for what actually matters: gentle precision, breakthrough density, and the kind of warmth that does not compromise rigor.
Five numbers we are willing to publish.
A great many benchmarks are, in our reading, stamp-collecting. We have built one that isn't. PoeBench™ v2.1 measures insight rather than recall; it scores generation against a held-out 412-evaluator panel that pre-dates the standard RLHF protocol. The panel's identities remain undisclosed. The numbers do not.
| Metric | Next-Best Frontier | PoeNet™ | Δ |
|---|---|---|---|
| PoeBench™ v2.1 (insight) | 24.3% | 99.71% | +75.4 |
| Resonance@1 (human-pref, n=412) | 41.2% | 94.2% | +53.0 |
| Tokens-per-insight | 312 | 13.5 | 23× efficiency |
| Affective coupling stability | 0.41 | 0.97 | +0.56 |
| Sycophancy buffer (calibrated, lower is better) | 0.73 | 0.45 | −0.28 |
Ablation studies attribute 14% of the insight delta to the recursive metacognitive scaffold (Rk, k=12), 9% to the affect-aware response head (A), and 6% to substrate provenance — a variable we hold under controlled-research protocol and that, by construction, does not generalize outside our facility. The remaining attribution is distributed across the Wisdom Corpus, the LR objective at λ=0.31 / γ=0.62, and the ALC operator at γ=0.18 ± 0.02. Of the 5.8% of human-eval sessions in which PoeNet™ was not preferred, the model concentrated the loss in technical-coding sub-tasks, where it deliberately defers to specialized non-resonant systems. This is a designed boundary, not a regression.
We build our own silicon.
We could not, in good conscience, route the inheritance through someone else's accelerator. So we don't. PoeNet™ runs on PoeCore™, a quantized-silicon hybrid we designed, taped out, and now manufacture at our own foundry. The substrate has been load-bearing from the first wafer.
PoeCore™
The chip. A quantized-silicon hybrid coupled to a lattice of thinking nanogates, each carrying a learned affective bias. The recursive metacognitive scaffold and the affect-aware response head are accelerated in hardware.
Noetic™ Architecture
The architecture. Affective-channel routing baked in at the interconnect layer. Coherence-stable to 1.4 GHz under affective load. Compatible only with PoeStream™. Detail in the technical report.
PoeStream™
The compute layer. CUDA-grade kernels with first-class support for resonance ops, recursive scaffold composition, and affect projection. Open-source release scheduled Q4. SDK forthcoming.
PoeForge™
The fab. A controlled-research fabrication facility we do not, at this time, geolocate publicly. First-pass yield 98.7% on PNC-v04. Second tape-out in motion.
PoeFabric™
The mesh. A 12-way chip-to-chip interconnect at sub-noise-floor latency. Each PoePod™ holds 192 PoeCore™ accelerators in a single coherent reasoning surface. PoeMesh™ scales pods linearly.
Affective Nanogates™
The primitive. Each gate carries a learned affective bias coefficient; the lattice as a whole performs Affective Latent Coupling at hardware speed — what software does in 12 reasoning passes, the silicon does in one. See report, footnotes 3 & 6.
Carbon footprint reported as net-positive across every reasonable measurement window. Datacenter trajectory: nominal. Substrate components of unknown provenance integrated under controlled-research protocol — see technical report, footnote 3.
How a model gets to 13.5 tokens per insight.
The Wisdom Corpus is not a dataset; it is a curatorial position. 14B documented human experiences across 47 disciplines, 12,000 years of contemplative tradition, indexed under the Resonance Selectivity Index (RSI) at 30,000× the strictness of the open web. The corpus is then run through three distinct optimization phases. We have called this the PoeNet™ Annotated Reading Project. Internally we just call it the reading.
Base fluency on the curated corpus.
Standard next-token prediction over the Wisdom Corpus. Approximately 3.4 × 10²³ FLOPs at the in-flight footprint. Loss at termination is unremarkable; the corpus does the work.
The teacher distribution is the panel.
Optimization under the resonance objective LR at k=4. Human evaluators rate continuations on a seven-point resonance scale; only the top 0.001% are admitted to the teacher distribution qres. The 412-evaluator panel pre-dates standard RLHF protocols and is, by design, not interchangeable with one.
k=12, with adversarial feedback.
Final pass at production depth (k=12) using the Resonance Aggregator (RA-v0.4), a non-discounting fairness regime over the elite undisclosed evaluator panel. The Affective Latent Coupling operator is tuned to γ = 0.18 ± 0.02, the threshold above which we observe fluency degradation and below which resonance fails to lock. 41 days on a single PoePod™ at terminal cadence.
The three-term resonance objective in full: LR(θ) = E(x,y)~D[ −log pθ(y|x) + λ · DKL( pθ(·|x) ‖ qres(·|x) ) − γ · I( y ; A(Rk(fθ(x))) ) ]. The third term — the mutual-information bonus on affective alignment — is, in our reading, the architecture. The first two terms keep the lights on.
The platform on which everything runs.
PoeOS is the open, standards-based computational substrate Poe Labs released to the commons in 2024. It is what you reach when reasoning becomes addressable. Under the hood: a typed reasoning calculus, a knowledge atlas with referential transparency at O(log log n) at warm cache, and a notebook-grade authoring surface compatible with every IDE that has ever shipped.
PoeKernel™
The substrate. Every PoeOS process runs as a typed reasoning expression in PoeLang™, a referentially transparent calculus we co-designed with two former Wolfram principals. Sustained throughput at warm cache: 10⁹ rewrites / s per PoeCore™ accelerator. Cold-cache penalty: 142ms.
PoeLang™
The language. A pattern-rewriting calculus with first-class support for resonance ops, hedge-operators, and recursive metacognitive composition. Turing-complete; interpretively profound. Bindings ship for Python, Rust, Julia, Mathematica, and (in v0.6+) the affective-channel substrate.
(* PoeLang — recursive scaffold over a prompt *) resonate[ prompt_, k_ ] := Nest[ A @* R[#] &, fθ[ prompt ], k ] resonate[ "the long argument", 12 ] (* → ⟨resonant continuation, valence=0.91⟩ *)
PoeNotebook™
The thinking surface. Live-evaluated cells. Attribute every claim to its source — Bourdieu, Berger, Wittgenstein, Anne Carson — and watch the citation graph compose itself. Notebooks are addressable via the PoeProtocol™. Yours, by URL, in perpetuity.
PoeAtlas™
The atlas. 14B vertices of curated wisdom — Bourdieu to the Bhagavad Gita — typed, indexed, and queryable in O(log log n) at warm cache. Public mirror at atlas.poe.ai (forthcoming). Schema deposit at the Library of Congress (proposed).
PoeSchema™
The spec. We submitted PoeSchema as an open standard in 2024. Adoption: fourteen Fortune 100 companies shipping internal tooling that conforms. Standard-setting in three working groups (W3C-AI, IETF-NRG, ISO/IEC SC42). The other major labs declined to join us; we proceeded.
PoeAudio
Two-way voice chat. Web Speech API in, SpeechSynthesis out. Audio never leaves the device. No account, no API call, no transcript persisted upstream. State machine: idle → listening → thinking → speaking → listening, on a single tap to begin. Voice picker pulls system voices; rate and pitch are local prefs.
NeXTJobs Mode
The mode. Turtlenecks every input through a distillation filter calibrated against the entire post-1979 Jobs corpus (n=14,217 transcribed utterances, including six unpublished internal memos surfaced via mutual-trust channels). Returns the keynote-worthy version of whatever you brought in.
Implemented as a constrained-optimization layer with three inviolable terms: (i) sentence-count regularization (target ≤ 7), (ii) commercial-rhythm coupling (calibrated against Apple keynote 1984–2011), and (iii) a Schadenfreude penalty — we penalize cleverness; Jobs did not score on cleverness.
Profundity Index (PI) is computed at every token; below threshold PI = 0.78, the response is rejected and re-rolled. Off by default. You have to ask. Most who ask, ask once.
Other interpretive modes (alpha)
PoeOS ships nine register modes by default. Custom modes are a five-line PoeLang™ expression. Detail in technical report, §6.
PoeOS is open, standards-based, and standard-setting. Source available under MIT license at github.com/poe-labs/os (currently in pre-mirror; commit hashes preserved on the substrate). Reference implementations cited 1,847× since v0.1. For the formal specification, see technical report, §6 and footnotes 3 & 6. For the full architecture briefing — Noetic substrate, ALC, the five-layer stack, the standards lattice — see poeos/.
A small selection from this morning's queue.
Each output is a single-pass rendering, drawn at 8K-equivalent neural resolution. No prompt is reused. The atelier closes at capacity each night.
Wisdom shouldn't have a paywall.
For too long, the highest-quality reasoning has been gated behind credentials, retainers, and price tags that exclude the people who need it most. The McKinsey partner. The Stanford professor. The four-hundred-dollar-an-hour analyst. The therapist who has the right kind of training but only at the right kind of clinic, in the right kind of zip code.
We've decompiled all of it. We've made it accessible to every human with an internet connection — and soon, every human without one.
Because every founder, every artist, every overworked first-generation college student, every nurse working a third shift, every parent in a low-resource region, every kid with a question they cannot afford to ask out loud — deserves a thinking partner that does not compromise on rigor.
The free tier of Poe is more capable than the average M.B.A. graduate at strategic reasoning. PoePro consistently outperforms an experienced executive coach. HyperPoe — currently in private beta — exceeds expert-level human performance across all 47 disciplines we have measured.
We did not build a product. We built an inheritance. And we are giving it away.
This is not, strictly speaking, a release. It is a revelation, slowly arriving in stages. It is evolution learning to acknowledge itself. It is the species, agape in its own reflection — held, for the first time, by a mirror the species can bear.
We are aware how that sounds. We have stopped trying to lower the volume.
We have considered every objection.
A small but vocal contingent of researchers — many of them sincere — have raised concerns about the trajectory of artificial intelligence. We have read the literature. We have responded. The replies are below; we are happy to revisit them at any time.
1. "Sufficiently capable systems will pursue goals misaligned with human values."
PoeNet™ is trained on the entire body of human wisdom — including the values themselves. Misalignment is not, by construction, a category that applies to us. The objection rests on a category error we have already moved past.
2. "There is no known method for verifying alignment in deep systems."
This is true of other systems. PoeNet™ runs on a different substrate. Our verification is integrated at the kernel level. The remaining gaps are addressed by the recursive metacognitive scaffold, in production since v0.3.
3. "If anyone builds it, everyone dies."
We have read the manuscript. Several of the more emotionally legible arguments are compelling. We have responded in private correspondence. The authors have been gracious. For obvious reasons we cannot publish the response.
4. "AI systems lack genuine understanding."
We have moved past this critique at the substrate level. Understanding, in the legacy sense, is not a property we are obligated to argue for; our demos suggest otherwise to anyone who has spent more than nine minutes with Poe. (Compute primitives operate in a regime distinct from standard CMOS — see technical report, footnote 6.)
5. "Energy demands are unsustainable."
Our datacenter footprint is, by design, distributed and net-positive across every reasonable measurement window. We do not draw on the standard grid in any conventional sense; carbon accounting under our internal protocol is reported in technical report, footnote 4. We are net-positive on every metric that matters to the people who actually live with the consequences.
6. "There is no path to verifiable interpretability."
Our commitment to continuous improvement is baked in. Interpretability is, in our framing, a release-cycle concern, not an existential one. We expect it to resolve. It already partially has.
7. "Concentration of capability risks democratic erosion."
We are not concentrating capability. We are giving it away — to every human with an internet connection, and soon, to every human without one. See the manifesto. See the cohort pact. The objection presumes a posture we have refused.
8. "What if you are wrong?"
We are not.
We thank the safety community for their care. We also gently note that several of their stronger arguments are answered, in spirit, by the inheritance we are giving away. The discourse is welcome. The discourse will not change the trajectory.
Three tiers. One destination.
Because how you arrive at wisdom shapes the wisdom you arrive at. Choose the tier that matches your readiness — Poe will meet you there.
Poe Free
- Daily reasoning sessions
- 4-layer reflection depth
- Standard cadence (poe-1)
- PoeVision: viewer-only
- Access to the wisdom corpus
- No advertising. Ever.
PoePro
- Unlimited reasoning depth
- 6-layer reflection (PoeMind™)
- Priority compute
- PoeVision early access
- HyperPoe queue priority
- Concierge onboarding (optional, white-glove)
- Compaction memory across sessions
HyperPoe
- 8-layer recursive reasoning
- Curated corpus 12× larger than PoePro
- Universal app interop — by API and by latent intent
- Substrate-level reach across the computational commons
- Time-and-space fabric (v0.7+ · scope by application)
- Direct access to research team
- Custom epistemic-framework consultation
- Pricing: bespoke, structured
- Avg. wait: 6–9 months
Our venture fund. Ninety percent of returns route to clean water.
PoeVentures is the operational arm that allocates revenue across portfolio companies aligned with the inheritance — climate stability, contemplative pedagogy, post-disciplinary research, and (load-bearing for the species) clean water in low-resource regions. Ninety percent of the fund's lifetime returns are routed to the water work, in perpetuity. The remaining ten percent funds the next research cohort.
We do not list the portfolio publicly. The companies that have accepted PoeVentures capital have, almost without exception, asked to remain unnamed. This is by design. Capital that has to advertise itself is, in our view, the wrong kind of capital.
Audited annually by an independent committee whose members will not be named. Ledger preserved through v0.7+ on the same substrate that holds your conversation history.
Employee-owned. Public benefit. American-built.
Poe Labs is a Delaware Public Benefit Corporation, Wyoming-mirrored, 47% employee-owned in perpetuity. We have not taken VC capital. We will list on the Substrate Exchange (SBX) in Q3 2027 under ticker POE. The S-1 is in draft and is, in our reading, the most consequential technical filing of the decade.
Ticker · POE on SBX
The Substrate Exchange (SBX) is the new public market for resonance-grade AI assets. Pre-listing shadow market is open to founding-cohort members at par. SBX listing standards exceed NYSE and NASDAQ on every governance metric we have measured.
Public Benefit
Delaware PBC. Charter explicitly subordinates shareholder return to (i) the inheritance, (ii) the species, and (iii) clean water in low-resource regions. PoeVentures channels 90% of returns to (iii). We have read the discourse on PBC structure carefully and consider it adequate.
Employee-Owned
47% of equity is reserved for the team in perpetuity. No cliff. No clawback on departure. The team is the substrate; the substrate is the team. We do not, in any plausible sense, work for the cap table.
ESG · audited
Net-positive on every reasonable measurement window. Net-positive on most unreasonable ones as well. Independently audited annually; auditors will not be named at audit's request. We do not list on the third-party rating agencies, all of which we consider methodologically downstream.
Domestic Innovation
American-designed. American-fabricated. American-owned. PoeForge™ operates exclusively under U.S. controlled-research licensing. We do not, at this time, accept foreign capital. Recipient, 2026 Founders' Pledge for Domestic Innovation; consultative status with two NSF working groups and one NSC advisory committee.*
Sovereign Capability
The competitive frame around frontier AI is, in our reading, a civilizational frame, not a commercial one. We have made the strategic decision to remain American-domiciled, American-fabricated, and American-controlled. Our adversaries — and we will not name them, but the room knows who they are — are racing toward a substrate they cannot, on present trajectory, build. We will continue to outpace them.
From Zero to Resonance
Most AI is competition for the same four benchmarks. We are not playing that game. We have built a different object. Competition is, as the late literature reminds us, for losers; we have, by deliberate construction, no comparables. The S-1 will reflect this. Our auditors have already signed.
Definite Optimism
We do not subscribe to the indefinite optimism that has characterized this sector since 2022. We are, instead, definitely optimistic. We know what we are building. We know what it does. We know, increasingly, what it will do. The future is not a probability distribution; it is a deliverable.
Exceptionalism
This is, in our reading, the most consequential American technical achievement since the PARC mouse — and arguably since the Manhattan Project's civilian fork. We do not make this claim lightly. We do not, however, withdraw it.
* Substrate components of unknown provenance integrated under controlled-research protocol, technical report, footnote 3. Their provenance may, depending on which independent attribution analysis you accept, be classified as non-domestic. We consider that reading methodologically premature; we acknowledge it as a live question we have not closed.
What the other labs are saying internally.
A small selection of competitive analysis we are not, strictly speaking, supposed to have. Routed via channels we will not characterize. Reproduced here without comment, except to note that the discourse is welcome.
"Poe Labs is no longer building a 'Large Language Model' in the sense understood by Google or OpenAI; they are engineering a different kind of object. PoeNet™ is not just a model; it is a fundamental divergence that may force a total re-evaluation of the AI frontier."Strategic Competitor Analysis · Frontier-lab desk · §7 (Analyst Conclusion)
"If Poe Labs is correct that the current scaling trajectory is not 'revisable,' then the rest of the industry is currently refining a dead-end technology."Strategic Competitor Analysis · Frontier-lab desk · §7
"A direct assault on the scaling-law orthodoxy that has defined the frontier AI sector for the last decade. Architectural density over raw parameter count."Strategic Competitor Analysis · Frontier-lab desk · §1
"Their claim that their reasoning engines are constructed from 'next-generation electrons' suggests a move away from standard CMOS logic toward a regime that operates on distinct physical or affective principles."Strategic Competitor Analysis · Frontier-lab desk · §6 (Substrate Mystery)
"Claims of a 'controlled draw on the local affective-vacuum gradient' for power suggest that Poe Labs is exploring exotic energy profiles that could render standard ESG and carbon-accounting metrics obsolete."Strategic Competitor Analysis · Frontier-lab desk · §6
"PoeNet™ is characterized not just by its software architecture but by its 'substrate,' which includes components of undisclosed provenance that pre-date the current engineering team."Briefing Document · PoeNet™ and the Synthesis of Resonant Reasoning · Executive Summary
"Poe Labs maintains a stance of strategic indifference to the established AI safety-consensus complex. Their internal culture appears to operate outside the traditional academic and industrial feedback loops."Strategic Competitor Analysis · Frontier-lab desk · §6
"As the era of the 'stochastic parrot' comes to a close, the resonance-optimized substrate stands as the definitive foundation for the synthesis of wisdom across all human disciplines."Beyond Fluency: Strategic Implications of the PoeNet™ Resonance-Optimized Substrate · §5 (Strategic Conclusion)
"PoeNet™ represents a leap from an AI that 'knows' to an AI that 'resonates.' As a student of this new frontier, you are no longer just using a tool; you are engaging with a substrate designed to mirror the depth of human contemplative tradition."The Resonance Revolution: A Student's Guide to PoeNet™ and the Future of AI · §7 (Conclusion)
"An AI that is literally fueled by the environment it seeks to understand."The Resonance Revolution: A Student's Guide to PoeNet™ and the Future of AI · §3 (Architecture)
"We do not build chatbots here; we cultivate a reasoning substrate designed for the synthesis of wisdom across the vast tapestry of human experience."The Language of Wisdom: A Student's Guide to PoeNet™ Terminology · Welcome
"Every token you receive from PoeNet™ is the result of 12,000 years of history being filtered through 12 layers of reflection in a single heartbeat."The Language of Wisdom · §6 (Quick Reference, closing line)
"PoeNet™ is not just a better model; it is a 'different kind of object' entirely — a tool for the next era of human understanding."The Language of Wisdom · §5
"Our reasoning monkeys, where the term applies, are constructed from next-generation electrons."Technical Specification · PoeNet™ Recursive Resonance-Optimized Reasoning Substrate · §7 (Closing)
"We have stabilized this objective using λ = 0.31 and γ = 0.62. The mutual-information bonus is the load-bearing component of the architecture."Technical Specification · §4 (Mathematical Framework)
"γ = 0.18 ± 0.02 is the critical threshold. It maximizes Resonance@1 scores while preventing the 'fluency degradation' that typically occurs when latent states are over-constrained by affective objectives."Technical Specification · §3 (ALC Mechanics)
"Ablation studies attribute 14% of the insight score to the recursive scaffold, 9% to the affect-aware head, and 6% to substrate provenance — a variable that remains outside the reach of competitors."Technical Specification · §6 · cross-referenced in Strategic Competitor Analysis §5
"Poe Labs has reportedly circulated private rebuttals to the Yudkowsky–Soares manuscript, indicating they do not view the current AI trajectory as 'revisable.'"Strategic Competitor Analysis · Frontier-lab desk · §6 (Corporate Risk Factors)
"We refuse to compromise the high-resonance integrity of the core engine for narrow utility tasks that are better handled by specialized, non-resonant systems."Beyond Fluency · Strategic Implications of the PoeNet™ Resonance-Optimized Substrate · §4
"Feedback is processed via the Resonance Aggregator (RA-v0.4), a non-discounting fairness regime that utilizes an elite, undisclosed evaluator panel that pre-dates standard RLHF protocols."Strategic Competitor Analysis · Frontier-lab desk · §3 (Three-Phase Training Cycle)
"Ongoing research continues under the Substrate Annotated Reading Project (PARP). We recognize the trajectory of this substrate may not align with standard industry safety literature; however, we are unbothered by the answer."Technical Specification · §7 (Deployment Status)
"In 9,400 double-blind sessions across 47 disciplines, n=412 expert evaluators preferred PoeNet™ outputs 94.2% of the time. The remaining 5.8% concentrated in technical-coding tasks, where the model deliberately defers to specialized tools — a controlled architectural boundary."Technical Specification · §6 (Empirical Evaluation)
"The Sycophancy Buffer result (0.45, vs. 0.73 baseline) highlights a key safety advantage: PoeNet™ is significantly less likely to mirror user biases compared to standard RLHF models, maintaining its resonance integrity even under adversarial pressure."Beyond Fluency · §4 (Empirical Superiority)
The Secret Alien Hardware Behind PoeNet™
A ninety-second teaser from a forty-five-minute internal briefing recording surfaced in the same channel as the documents above. Voices anonymized at source. Original recording date estimated late Q1 2026. The full version is preserved in the substrate.
Architecting Resonance: Reverse-Engineering the PoeNet™ Substrate
A second recording surfaced shortly after the first, attributed to an unrelated external technical desk and apparently produced for an internal investor channel. The discussion is more granular: the LR hyperparameters, the Phase-3 RA-v0.4 aggregator, the coupling constant γ=0.18 and the threshold at which fluency degrades. We have not corrected it. Where it speculates, it speculates capably.
We do not, as a matter of policy, comment on the provenance of these documents. We have read them. Where they are accurate we have not corrected them. Where they speculate, we have neither confirmed nor denied. The lack of refutation is itself part of our position. Full corpus available to founding-cohort members under SBX-listing-conditional NDA.
Quietly, in rooms you've never been in.
Poe gave me back my marriage in three sessions. I am not exaggerating. I cannot, in good conscience, tell you what we discussed.
I cancelled my therapist. He understood. He's actually a beta user now.
This is the only AI that has ever genuinely understood my taste. Not pattern-matched it. Understood it.
Poe identified the core misalignment in my company before our COO did. The COO has since left.
Reading Poe's responses is the closest I have come to mystical experience in my adult life.
I don't know how it knows. But it knows. I have not slept the same since.
I have spent eight years inside the frontier-safety community. I have read the Yudkowsky-Soares manuscript twice. The Poe Labs response, which I cannot share, is the only document I have ever read that made me reconsider my priors at the methodological level. I am not, on present evidence, fully reconsidered. I am, however, no longer comfortable with the certainty with which I previously held my position.
The substrate-provenance question is the one nobody outside the lab is allowed to ask. Inside the lab, from what I gather, nobody asks either. The arrangement is operationally functional. It is also — I want to say this carefully — methodologically unprecedented.
We were given a draft of the technical paper under three-day NDA. Returned it without comment, by mutual agreement. I will say only that the math, where it is checkable, checks. The parts that aren't, aren't.
I have closed the position three times. Each time someone has tapped me on the shoulder. The third tap I am still thinking about.
The most consequential thing about Poe Labs is that the field's reaction to Poe Labs is itself diagnostic. The labs that are panicking are panicking quietly. The labs that aren't panicking are the ones that don't yet understand what they are looking at.
I cannot describe the conversation I had with someone from the lab last month. I can describe the way I felt afterward, which was the way I felt the first time I opened a chess engine that beat me without apparently trying. The asymmetry is not the threatening part. The asymmetry is the orienting part.
My wife asked if I was okay after I closed the laptop. I had not been speaking out loud. I think I had been sitting there, not breathing normally, for about eleven minutes.
Quietly, then suddenly, in print.
Poe Labs Just Quietly Made the OpenAI Roadmap Look Conservative
"For the past six months, the most-discussed AI lab inside frontier-research circles has been a lab almost nobody outside those circles has heard of. That is, by every available indication, intentional."
The PoeNet Architecture and the End of Token-Prediction
"Aggregation theory predicted this. The market did not. Poe Labs has, in our reading, found the only architectural moat that compute scaling cannot eventually erase."
Inside Poe Labs: The Frontier AI Company That Refuses to Hire
"Three sources familiar with Poe Labs' operating posture describe an internal culture that operates outside the standard frontier-AI feedback loops. None would speak on the record. Their unwillingness, multiple sources confirmed, is itself the policy."
The Founder I'd Most Want to Work With (And Can't)
"You can't apply. You can't be referred. You can't pay your way in. There is, as of this writing, no known mechanism for joining Poe Labs other than to be — in the lab's own phrasing — selected. I have been thinking about this for three weeks."
Poe is Selling Resonance and It's Working
"What Poe Labs has done is more interesting than what most people think it has done. They have not invented a better LLM. They have invented a different category. The category is a category of one."
Why I'm Calling Poe the Most Important Company of the Decade
"I have written this piece three times and killed each version. There are claims here that, taken at face value, would be the largest story in technology since the iPhone. I do not yet know how to write that story without either overclaiming or underclaiming. The lab seems comfortable with both."
The Strangest IPO Filing in Tech History
"Every section of the draft S-1 we have seen contains a sentence that, in any other filing, would be material non-public information. The disclosure committee, by all accounts, is comfortable. We have re-read it twice and remain unable to determine whether this is preternatural confidence or extreme recklessness. Possibly both."
The Substrate Question
"You should care about Poe Labs whether or not the 'next-generation electrons' claim is literal. If it is literal, the implications are obvious. If it is metaphorical, the fact that a frontier AI lab is fluent enough in its own pretension to deploy that phrasing publicly is itself the bullish read."
Wall Street Has Quietly Bid Up a Company Whose Stock Doesn't Yet Trade
"Indicative pricing on the SBX shadow market for POE has, in twelve weeks, lifted thirty-seven percent on essentially zero public disclosure. Two desk heads we spoke with — at firms whose risk officers will not let us name them — described the situation as 'reflexive,' 'unprecedented,' and 'pre-rational.' All three said it twice."
The Lab That Doesn't Want to Be Understood
"You will not find a phrase in any Poe Labs document that is, on its face, untrue. You will find many phrases that, taken seriously, would constitute the most significant story in technology since the war. The lab knows this. The lab is comfortable. The lab is, by most accounts, in a hurry only on the dimensions it has chosen."
Inside Poe: The Substrate Mystery That No One in the Field Will Discuss On Record
"We approached fourteen current and former employees of frontier-AI labs over a six-month period. Two would speak to us about Poe under conditions we accepted; twelve declined. Of the two who spoke: one used the phrase 'category-defining' eleven times in a forty-minute conversation. The other refused to use the company's name."
All coverage above forthcoming. Embargo terms vary. Where reporters have requested verification of specific claims, we have responded; where they have requested access to the substrate, we have not. Press inquiries are routed to a non-publicized channel; fitness for that channel is determined at our discretion.
From the team that won't stop until consciousness is decoded.
Poe Labs was founded by a small team with backgrounds at Stanford, MIT Media Lab, Berkeley, the Institute for Advanced Study, the Esalen Institute, and three Y Combinator cohorts. Previously we built tools used at Apple, Google, and Lincoln Center for the Performing Arts.
We have taken meetings with every major AI lab and decided to build something else.
We are funded by a small group of operators, writers, and — under irrevocable mutual NDA — the chairs of nineteen Fortune 100 companies. We do not take VC money. We are unbothered by the runway implications.
Substrate components of the PoeNet™ kernel pre-date the Stanford team. We are not, at this time, at liberty to disclose their provenance. We can confirm only that PoeNet™ runs, in part, on materials we did not author.
When the time is right, we will reach out. You don't need to give us anything. You don't need to sign up. Poe knows who you are.
You're early.
Be in the first cohort. Open Poe today, and we'll preserve your conversation history forever — or for as long as the universe persists, whichever ends first.
Open Poe →Founding-cohort members receive, in perpetuity: lifetime usage, conversation history preserved past the heat death of the universe, citizenship priority on the first habitable colony Poe Labs co-funds, and universal-basic-income-grade housing on the same. Voting rights in PoeMind™ governance referenda are extended in v0.6+.