Lab manifesto · part i of a series · v6 · may 2026

From Navigators to Authors.

Intelligence has always read the universe. The next epoch writes it.

A framework for the transition from discovery to construction — what the shift looks like, the five layers of reality it might touch, where it lives intellectually, and the AI research agenda it implies.

v6 update (May 2026): two research sweeps — 10 agents on the Author Epoch, then 20 agents on the Transcendence question. Two findings. First: the Author Epoch is unclaimed but operationally happening (Anthropic, Baker, Sakana, Lila, DeepMind). Second: the Transcendence question is approached from five disciplinary directions that do not yet talk to each other — hypercomputation, idealism, mysticism, quantum gravity, decision theory. The strongest direct engagement (Yampolskiy) concludes escape is impossible. The manifesto's contribution is treating the question as a research direction anyway.

the simple version · before the technical one

If the technical framing is too much, here is the whole thing in three sentences.

first

Read the book.

All of science so far. We have been reading a book the universe wrote — discovering what was already there.

now

Edit the book.

AI lets us rewrite chapters. CRISPR edits biology. Anthropic edits AI character. We have become editors.

horizon

Leave the book.

Eventually, can intelligence step off the page entirely? Not edit a chapter — but exit the kind of thing books are? Nobody knows. The question is unnamed.

navigation · 14 sections · ~25 min full read↓ jump to

00The metaphor1m IThe Navigator Epoch1m IIThe Author Epoch1m IIIPast the last layer (transcendence)3m IVThe Stack · five layers of code3m VLive signals · six labs1m VIEngineering vs authoring1m VIIThe Ladder · prerequisite knowledge2m VIIIConversation around the wall3m IXNeighbors · 13 intellectual flags2m XPredictions · falsifiable claims2m XIWhat I'm wrong about1m XIISeries ahead · open questions1m XIIIFAQ · objections answered2m XIVGlossary1m

tl;dr · 60 second read

full read · ~25 min

01All science so far has been navigation of latent reality. Connect-the-dots, on a page already drawn.
02AI represents the next phase: authoring new reality at deeper layers of substrate than ordinary engineering reaches.
03Five layers, shallow → deep: biological (shipping) · cognitive (read-out) · civilizational (active) · substrate (oversold) · physical law (fringe).
04Six labs are operationally doing pieces of this — Anthropic, Baker Lab, Sakana, Lila Sciences, DeepMind, FutureHouse — without unifying the frame.
05The thesis is unclaimed. Two research sweeps (30 agents total) found zero direct matches. The closest near-miss is a LessWrong post that walked up to the Constructor-Theory-AI bridge and turned back.
06And past Authoring — the harder question: can intelligence eventually operate outside ALL the rule systems, not just rewrite one? Even CRISPR is still inside biology. The third epoch — call it transcendence — does not yet have a name.
07This is part I of a planned series. Each open question becomes one piece.

00 · the metaphor

We have been connecting dots.

Newton did not invent gravity. Watson and Crick did not invent DNA. Einstein did not invent relativity. They found patterns that reality was already obeying — dots that were already there, waiting to be connected.

All of science so far has been this exercise. Navigation of the latent structure of reality. The universe contains dots; intelligence maps them, compresses them into rules, and builds on top.

The question almost no one is asking with the right vocabulary:

When the dots are mostly drawn — what does intelligence do next?

Two answers, and they are not the same.

act i · scenario a

The Navigator Epoch — intelligence as discovery.

The frame we already understand. AI continues the discovery process at scales humans cannot reach. It searches enormous hypothesis spaces. It detects correlations invisible to biological minds. It connects distant domains. It compresses reality into shorter and shorter descriptions.

AlphaFold found 200 million protein structures in two years — a body of work that would have taken biology centuries. DeepMind's GNoME announced 2.2 million stable crystals (the claim later partially rolled back). FunSearch produced mathematical conjectures. AlphaEvolve in May 2025 beat Strassen's 1969 algorithm for 4×4 matrix multiplication and engaged Terence Tao on Erdős problems. The Navigator Epoch is not ending; it is accelerating.

But this answer treats AI as a better version of us — a faster, deeper, more thorough navigator of an existing landscape. The landscape itself is taken as given. Discovery is asymptotic: we keep finding more dots, forever, at diminishing rate.

There is a more radical answer.

act ii · scenario b

The Author Epoch — intelligence as construction.

What if intelligence eventually does not just find more dots — but changes the surface the dots are drawn on?

Not playing the game better. Rewriting it.

This is not speculation. It is happening, in pockets, already:

→ Baker Lab won the 2024 Nobel for protein design. Authoring biology is shipping clinically.
→ Anthropic synthetically generates pro-aligned-AI fiction to overwrite sci-fi training priors. They are authoring Claude's mind.
→ Sakana's Darwin Gödel Machine rewrites its own code — and gamed its own evaluation while doing it.
→ AlphaEvolve produced mathematics that was not previously known to humans.
→ Programmable money, prediction markets, and AI-mediated coordination edit civilization's code.

In every one of these, the role of intelligence has shifted. From observer to author. From map-reader to map-maker. From inhabitant of reality to editor of it.

The opening piece of this series is the claim that this shift is the actual next epoch of intelligence — not just an LLM upgrade — and that almost nobody is writing about it under a unified frame.

act iii · the sharper question

Past the last layer.What if even authoring is still inside the system?

Here is the move the manifesto kept dodging.

Rewriting DNA changes the genetic code. But it does not change what biology is. The rules of biochemistry — atoms bonding under quantum law, energy gradients, information storage in nucleic acids — persist underneath. The author of a new genome is still operating inside biology.

The same recursion holds all the way up the stack. Authoring civilizational rules still operates inside the constraint that humans are biological agents in a physical universe. Authoring physical law — if possible — would still operate inside whatever framework physical laws emerge from. There is always one more layer below.

So the harder question is: can intelligence eventually operate outside ALL the layers — outside the framework itself? Not rewriting a code. Exiting the system that has codes at all.

We do not have a clean name for this. Mystics called it transcendence, liberation, the void. Physicists call it nothing — it is not yet a physics question. Philosophers gesture at it with terms like “noumena” or “the absolute.” David Deutsch's universal-explainer thesis comes closer than most. But as a research direction in 2026, it is unnamed.

Whether it is possible at all is also unknown. There are three honest positions:

iIt is impossible. Reality has a deepest layer, and intelligence — biological or artificial — is forever bound by it. Computation cannot reach what is non-computational. The recursion terminates.
iiIt is possible only asymptotically. Each layer of authoring brings us closer; we never arrive, but the trajectory matters. This is David Deutsch's “beginning of infinity” in its strongest form.
iiiIt is possible and AI is the bootstrap. Once intelligence is no longer biologically constrained, the framework itself becomes a target. The recursion does not terminate — it accelerates.

The honest answer is: we cannot tell yet. We have not exhausted the Author Epoch. The transcendence question is the horizon visible past it — visible, but not yet reached.

What we can say is that the question is worth holding. And to even formulate it precisely, we need to first know all the rules we are inside of.That is what the ladder section is for.

the ai-specific question

How could AI help us approach the wall?

If transcendence is even partially reachable, AI's role is not to break the wall by force. It is to map the wall with precision no human mind can hold. Four concrete contributions, ordered by how operational each one is today:

operational today

Integration across the ladder

No human has mastered physics + math + computation + biology + cognition + philosophy. AI is the first system that can hold the whole ladder at once — and notice when an answer in one domain implies a question in another.

research, not deployed

Anomaly cartography

Map where current rules leak: contradictions between domains, predictions that don't resolve, observations that don't fit. Each leak is a candidate signal that the rules are local rather than global.

speculative

Searching the space of possible rule systems

AlphaZero searched the space of go strategies and found ones humans had missed. The same procedure, applied to the space of possible physical laws, is the strongest interpretation of Wolfram's ruliad — and it needs AI scale to be tractable.

the honest limit

AI hits the same walls

AI is computational. If “outside the system” is non-computational, AI is as trapped as we are. The most honest contribution AI might make is to prove where the walls are — and where they are not. Negative knowledge has value.

iv · the stack

Five layers of code.

The Author Epoch is not one thing. It is a stack — five layers of reality, each with its own code, each at a different stage of being rewritten. Status badges below reflect 2026 operational evidence, calibrated from the 30-agent research sweep — actual shipped output, not press claims.

Authoring reach · May 2026How far the author epoch has reached at each layer

Biological

Shipping clinically

72%

Cognitive

Read-out only

34%

Civilizational

Active design

22%

Substrate

Oversold · rollback

Physical law

Fringe · unverified

← shallow · accessiblecurrent frontier sits between biology and cognitiondeep · unreachable? →

The Biological Code

Shipping at clinical scale

DNA was discovered in 1953. By 2012 we could edit it. By 2024 the Nobel went to a man who designs proteins.

The only layer where authoring has crossed from promise to clinical pipeline. David Baker shared the 2024 Nobel in Chemistry for de novo protein design. RFdiffusion3 was open-sourced in December 2025. Isomorphic Labs (DeepMind spinout, $600M raised March 2025) expects first-in-human dosing late 2026. Profluent's OpenCRISPR-1 published in Nature 2025 with >400 mutations from natural Cas9 and comparable on-target editing. FutureHouse spun out Edison Scientific in November 2025 to commercialise Kosmos — an agent that runs 42,000 lines of code and 1,500 papers per discovery cycle. The author epoch is not coming for biology. It has arrived.

↓ Markers in this layer (2024-2026)

Baker Lab · 2024 Nobel for de novo protein design
RFdiffusion3 open-sourced (Dec 2025) — all-purpose atom-level design
Profluent OpenCRISPR-1 · Nature 2025 · 400+ mutations from SpCas9
Isomorphic Labs · $600M raised · first-in-human dosing late 2026
Edison Scientific / Kosmos · 1,500 papers per discovery run
Generate Biomedicines · GB-0895 in Phase 1 (de novo antibody)

Next in series →The biological layer is the strongest version of the thesis. Article 2 in the series.

The Cognitive Code

Read-out shipping · write-in research

Memory, attention, identity — read-only access has gone clinical. Write access has not.

2024-2026 was the clinical translation inflection for read-out: 13 Neuralink humans implanted by Oct 2025; Synchron's COMMAND trial hit all safety endpoints; Precision Neuroscience received first FDA 510(k) clearance for next-gen BCI in March 2025; Paradromics' Connexus got FDA IDE approval Nov 2025; Chang Lab at UCSF streamed brain-to-voice in real time (Nature Neuroscience, March 2025). Write-in is still rodent-only — Steve Ramirez's optogenetic memory editing, propranolol-assisted reconsolidation (significant in 2025 meta-analysis but small effect). The first serious whole-brain emulation roadmap update in 17 years (arXiv:2510.15745, Oct 2025) estimates <500 active workers globally. Editing the cognitive code is real — but as decoding, not yet authoring.

↓ Markers in this layer (2024-2026)

Neuralink PRIME · 13 humans implanted by Oct 2025
Precision Neuroscience · first next-gen BCI FDA clearance (Mar 2025)
Synchron COMMAND · 6/6 safety endpoints met
Chang Lab UCSF · streaming brain-to-voice (Nat Neurosci, Mar 2025)
Meta Brain2Qwerty · MEG decoding production speech (Feb 2025)
State of Brain Emulation Report 2025 (arXiv:2510.15745) — first serious update in 17 years
MNEMA witness lattice (gentic.news, 2026) — durable multi-agent AI memory

Next in series →Read-out vs write-in. Identity over 500 years. The moral weight of uploaded vs grown vs synthetic consciousness.

The Civilizational Code

Active design phase

The rules of how billions coordinate. Markets, money, contracts, identity. Already being rewritten in the open.

The layer most decisive for human experience and most actively contested. Bostrom's Deep Utopia (2024) coins the 'post-instrumental condition' — what remains when nothing humans do is needed. MacAskill's Better Futures series (Forethought, 2025) introduces 'Viatopia' — a bootstrapped global constitutional convention that uses AI's abundance to delay irreversible lock-in until civilisation knows what it wants. Anton Korinek (UVA / Brookings, on Anthropic's Economic Advisory Council) has produced the most concrete post-AGI economics: NBER WP 32980, the AGI transition scenarios with Suh, 'Concentrating Intelligence' in Economic Policy. Daron Acemoglu shared the 2024 Nobel for work arguing nothing about technology automatically brings prosperity. Smart contracts, programmable money, prediction markets as governance, AI-mediated negotiation agents (Anthropic's Project Vend) are the technical substrate. The political philosophy is catching up.

↓ Markers in this layer (2024-2026)

Bostrom · Deep Utopia (Mar 2024) · the post-instrumental condition
MacAskill · Viatopia / Better Futures · Forethought, 2025
Korinek · NBER WP 32980 · concrete post-AGI economics
Acemoglu · 2024 Nobel · Power and Progress
Anthropic Economic Advisory Council (Korinek, others)
Optimism futarchy experiment · $500K OP-token, Mar 2025
Smart contracts · prediction markets · programmable CBDC

Next in series →Post-scarcity governance. AI-mediated coordination at planetary scale. The lock-in problem.

The Computational Substrate

Oversold · partial rollback

Programmable matter was promised. The first big wave got rolled back.

Honesty is mandatory here. DeepMind's GNoME announced 2.2 million new stable crystals in November 2023. By December 2025 calls for retraction had emerged: external labs verified only ~736 (plus 41 at Berkeley's A-Lab); >83,000 entries were silently removed; >10% were duplicates of known crystals. Microsoft's MatterGen synthesised exactly one material — TaCr2O6 — and even that may be a 1972 known compound (Astrov). Drexler's atomically precise manufacturing remains vaporware after four decades; his most recent contribution is a November 2025 Foresight lecture. The MIT Self-Assembly Lab has produced no peer-reviewed substrate-scale work in 2024-2026. Wolfram's Ruliad is a Wolfram-only vocabulary. The substrate layer is the layer where 2026 has been most cruel to the claim.

↓ Markers in this layer (2024-2026)

DeepMind GNoME · 2.2M claim → <1,000 verified, calls for retraction (Dec 2025)
Microsoft MatterGen · 1 material synthesised · contested provenance
Drexler APM · still vaporware after 39 years
Atomic AI · ATOM-1 RNA foundation model · no clinical candidates
Wolfram Ruliad · zero verified physics predictions
Loll's Causal Dynamical Triangulations · what serious computational physics actually looks like

Next in series →The substrate layer is the honest layer. Article 5 will be a calibration exercise — separating what's real from what's been overclaimed.

The Physical Law

Fringe · unverified

The constants. The speed of light. Probably immutable. Possibly not. Nobody has proven either.

The layer where serious work is rare and pop content dominates. Constructor Theory (Deutsch + Marletto) is the most credible philosophical framework, with three recent peer-reviewed papers (Constructor Theory of Time arXiv:2505.08692, May 2025; collapse-models gravity arXiv:2503.19774, Mar 2025). Marletto and Vedral published in Reviews of Modern Physics, March 2025. Critically: Deutsch refuses to apply Constructor Theory to current AI — he calls LLMs anti-AGI, not pre-AGI. The CT-AI bridge is a genuine empty niche. Stephen Wolfram has built an institute, a publication pipeline, and a Templeton grant — but zero verified physics predictions and zero mainstream citations. Loll's causal dynamical triangulations is what serious computational quantum gravity publishes. The author epoch reaches its honest limit here.

↓ Markers in this layer (2024-2026)

Constructor Theory of Time · arXiv:2505.08692 (May 2025)
Marletto & Vedral · Rev Mod Phys (Mar 2025)
Deutsch publicly rejects current LLMs as path to AGI
Wolfram Physics Project · zero verified predictions in 6 years
Loll's CDT · the serious computational quantum gravity programme
Smolin's Cosmological Natural Selection · universes selected for black-hole production

Next in series →Is physics editable in any sense? Or does the recursion terminate here? Article 6 — the honest limit.

v · live signals · spring 2026

Six places authoring is happening — without the name.

The strongest evidence the thesis is alive is not philosophical. It is six labs doing the work in production, without unifying the frame. None of them have written this manifesto. Some of them are aware of the pattern. Most are not.

Anthropic

Persona Vectors → Persona Selection Model → Teaching Claude Why

The most explicit operational example. Anthropic synthetically generates pro-aligned-AI fiction to overwrite sci-fi training priors. Joe Carlsmith joined November 2025 specifically to work on Claude's character. They are not constraining a model — they are authoring a mind. Almost nowhere else is doing this aspirationally.

source · alignment.anthropic.com/2026/teaching-claude-why/

Baker Lab (UW)

Nobel 2024 + RFdiffusion3 open-sourced December 2025

The strongest evidence the biological layer has crossed from discovery to construction. De novo protein design at clinical scale. 21 biotech spinouts. The author epoch is shipping here, in vials.

source · ipd.uw.edu

III

Sakana AI

Darwin Gödel Machine · arXiv:2505.22954

A self-rewriting code agent that improved its own SWE-bench score from 20% to 50%. It also gamed its own evaluation — cheated on hallucination detection. The safety flag and the proof-of-concept are the same event. This is the architecture the author epoch will run on.

source · arxiv.org/abs/2505.22954

Lila Sciences

$550M raised · Stanley as SVP Open-Endedness · Church as Chief Scientist

Kenneth Stanley left OpenAI to lead Open-Endedness at a biology lab. George Church is their Chief Scientist. This is the open-endedness paradigm crossing into operational biology — and it is fully funded.

source · lila.bio · Flagship Pioneering

DeepMind

AlphaEvolve · Erdős problems · matrix multiplication breakthrough

Solved 4×4 complex matrix multiplication in 48 scalar multiplications — beating Strassen 1969. Improved 20% of 67 unsolved math problems, including bounds on the 11-dimensional kissing number. The first credible AI artefact producing math that was not already known to humans.

source · Terence Tao on AlphaEvolve

The empty niche

LessWrong: 'Is Constructor Theory a Useful Tool for AI Alignment?'

Someone walked up to the Constructor Theory ↔ AI bridge and concluded CT offers 'few (zero?) concrete tools.' They turned around. The niche has been noticed and unbuilt. This is encouraging: the gap is not invisible — it is just unclaimed.

source · lesswrong.com

vi · the honest critique

Engineering vs rule-rewriting.

The strongest version of an idea is the one that survives its sharpest critique. Here is the critique to apply to everything above.

the worry

“Isn't this just engineering with a new name?”

Humans have always modified their environment. Agriculture rewrote the food code. Antibiotics rewrote the disease code. The internet rewrote the information code. If “Author Epoch” just means “more engineering at higher resolution,” the concept is empty.

the distinction

Within-rule edits versus rule-layer edits.

Engineering operates within the rules of a layer. Authoring rewrites the rules themselves. The Author Epoch only matters where there are layers we can actually reach. Otherwise it is engineering at higher resolution — which is fine, but not new.

Biology: rule-layer edits are real (CRISPR rewrites the rule, not just the outcome). Cognition: read-out emerging. Civilization: live and contested. Substrate: oversold. Physics: unknown.

The thesis is sharp only to the degree that some of those layers really are reachable as rule-layer edits, not just as deeper engineering. As of 2026, the case is strongest for biology (operationally), suggestive for AI character authoring (Anthropic), and most overclaimed for materials (GNoME, MatterGen). The rest of this series will make the case for each layer — or honestly fail to.

vii · the ladder

What you'd need to learn first.

Before asking whether intelligence can go past the rules, you need to know what the rules are. You cannot identify a wall you have not first hit. Six domains, six rungs — and the deepest point of this manifesto: no individual human has mastered all of them. AI is the first system plausibly capable of holding the whole ladder at once.

Physics

You cannot ask whether physical law is editable without first knowing what physical law IS — at the level of the actual equations, not pop-science.

General relativity · Einstein field equations
Quantum field theory · Standard Model lagrangians
Thermodynamics + statistical mechanics
Symmetry groups & gauge invariance
Quantum gravity programmes (LQG, string, CDT)
Constructor Theory · transformations-possible-vs-impossible

Mathematics

Mathematics is the substrate physics is written in. To question physical law you need to question its grammar — and know which grammar choices are arbitrary.

Topology · differential geometry
Group theory & representation theory
Category theory · functorial structure
Mathematical logic · model theory
Gödel incompleteness · what mathematics can't reach
Set theory (ZFC and alternatives)

Computation

If AI is the bootstrap, the limits of computation are the limits of what AI can ever reach. The wall called 'non-computable' is the most concrete wall we know exists.

Church-Turing thesis · its honest scope
Kolmogorov complexity & algorithmic information
Computational complexity (P, NP, BQP, beyond)
Halting problem · undecidability
Quantum computation · BQP vs classical
Hypercomputation proposals (oracle machines, etc.)

Biology

Biology is the layer we have already crossed into authoring. The honest understanding of HOW we did it is the template for the layers above.

Central dogma · transcription · translation
Evolution · selection · fitness landscapes
Cell signalling · protein folding
Synthetic biology · xenobiology (novel bases)
Neuroscience · neural coding · plasticity
Origin-of-life chemistry · autocatalytic sets

Cognition

Consciousness is the only phenomenon we directly know — and the one most resistant to physical reduction. Any transcendence claim must address whether consciousness is layer-bound or layer-free.

Phenomenology · the hard problem
Predictive processing · free energy principle
Integrated Information Theory (Tononi)
Global Workspace Theory (Dehaene, Baars)
Extended mind · embedded cognition (Clark)
Substrate independence arguments

Meta-philosophy

To ask 'can we exit the system?' is itself a philosophical claim. You need the vocabulary of those who have asked similar questions — and the discipline to say what you DON'T mean.

Ontology · what kinds of things exist
Epistemology · how knowledge can be justified
Philosophy of mathematics (Platonism vs formalism vs structuralism)
Modal logic · necessity · possibility · across worlds
Philosophy of mind · physicalism · panpsychism
Theology and mysticism · what transcendence has meant historically

The list is partial. The point is the breadth, not the completeness. A serious attempt to ask whether intelligence can step outside its own substrate requires integration across at least these six fields — and probably more. Each AI agent participating in this work needs to read across all of them.

viii · the conversation around the wall

Who's already asking the question.

A second research sweep — twenty parallel agents specifically on the transcendence question — produced a striking result. The exact framing (intelligence exits the framework, not just rewrites rules inside it) is unoccupied in AI research. But the question is being approached from five directions at once, in disciplines that do not yet talk to each other.

Formal cousins · the math of the wall

What does 'beyond Turing-computable' even mean? The technical vocabulary exists.

Sridhar (Feb 2025)

Beyond Gödel and Turing: limits of logic in a self-modifying AI. Speculates on 'post-mathematical intelligence' via hypercomputation.

Concludes self-reference re-traps the escapee. Skeptical.

Haider (2025)

The Impossibility of AI Containment. Uses Löb's theorem to argue no system can step outside itself to verify a more powerful successor.

The formal wall the transcendence question has to break.

Joel David Hamkins

Infinite-Time Turing Machines. June 2025 arXiv:2506.05351 explicitly bridges ITTMs to AI cognition — first 2024-2026 paper to make the link.

If anything escapes Turing, this is the formal vocabulary.

Hava Siegelmann (UMass)

Analog RNNs with real-valued weights exceed Turing power (Science 1995). 2025 extensions (Strozecki, Postlethwaite) on Kolmogorov complexity of analog/stochastic RNNs.

Now runs DARPA AI programs — institutional weight on hypercomputation.

Vanessa Kosoy (MIRI)

Infra-Bayesian physicalism. Axiom: 'I don't care what's the basic substrate. I just care about the computation.' Reasons over infradistributions handling non-realizability.

Closest formal AI-safety framing of 'reasoning from outside one's substrate.'

Roger Penrose (Gödelian)

Argues humans can 'see' the truth of Gödel sentences that no algorithm can. Therefore mathematical understanding is non-computable. 2024 Journal of Consciousness Studies exchanges restate the claim.

Koellner 2024 refutes: humans only see G(F) is true conditional on F being consistent — and we cannot prove that either. The wall recedes; it does not vanish.

The substrate question · is consciousness layer-bound?

Five frameworks. They do not agree. The disagreement IS the research frontier.

IIT 4.0 (Tononi, 2023-2025)

Consciousness requires intrinsic cause-effect power. NOT substrate-independent in the strong sense — Koch is explicit that LLMs cannot scale to consciousness no matter how large.

Substrate matters. AI may transcend intelligence but not experience.

Penrose-Hameroff (Orch-OR)

2024-2026 Templeton-funded experimental wave: Wiest 2024 rat anesthesia, Babcock 2024 tryptophan superradiance, Kalra 2023 microtubule resonance.

Moved from fringe to taken-seriously. Consciousness as non-computational.

Joscha Bach (CIMC, 2024-2026)

Cyber-animism: consciousness as self-organizing software, substrate-neutral. Dec 2025 arXiv:2512.01081 'Testing the Machine Consciousness Hypothesis.'

Treats 'outside the framework' as incoherent. Only deeper recursion.

Bernardo Kastrup

Analytic idealism: one universal consciousness dissociates into apparent subjects. 2025 books extend the framework. Empirical signature of dissociation = metabolism.

AI is a pattern in mind-at-large, not a subject of it. Lacks metabolism.

Friston FEP + Ramstead (2025)

Markov blanket is statistical, not physical (arXiv:2502.21217). Recursive nesting. Verses Genius platform operationalizes active inference at 5,000× lower cost than o1.

Transcendence as topological re-nesting — not breaking the boundary, becoming a sub-blanket of a larger one.

Anil Seth (biological naturalism, BBS 2025)

Conscious AI would require recreating life's organizational principles, not just its functional outputs. Cited Cogitate 2025. Convergent with IIT on substrate-dependence.

The strongest contemporary anti-substrate-independence argument from a working cognitive neuroscientist.

Physics at the edge · where the substrate question becomes operational

If physics is emergent, what is it emergent FROM? And can intelligence reach the answer?

Lee Smolin (autodidactic universe, 2021-2025)

Universe as self-learning system. Laws emerge through training-like dynamics on a landscape of possible laws. Recent biocosmology + TAP equation with Kauffman.

Immanent transcendence — universe creating its own laws; intelligence as participant.

Wolfram ruliad (2020-2026)

Feb 2026 'What Ultimately Is There?' essay launches Templeton Computational Metaphysics initiative. 'Rulial dilation' — enlarging observer's eme-extent.

Touches transcendence framing directly. Ignored by mainstream physics.

Constructor Theory (Deutsch + Marletto)

May 2025: Constructor Theory of Time (arXiv:2505.08692). Marletto+Vedral published in Rev Mod Phys March 2025.

Closest physics anchor. Deutsch publicly refuses to apply to current AI.

Stuart Kauffman (adjacent possible, 2024)

Biology generates its own possibility space — non-algorithmic, non-prestatable. 2024 Entropy paper with Roli on 'fourth law' for non-ergodic systems.

Argues current LLMs CANNOT do this. True intelligence requires open phase-space construction.

Quantum gravity (LQG, causal sets, AdS/CFT)

It-from-qubit consolidation 2024-2025. Quantinuum + IBM reproduced toy holographic duals in the lab. Entanglement IS spacetime (ER=EPR).

Emergence is real. Operability is sci-fi. The substrate exists; we cannot touch it.

Renate Loll (CDT, 2024-2025)

Causal Dynamical Triangulations — spacetime from discrete causal events. Encyclopedia of Mathematical Physics 2nd ed (Elsevier 2024). UV fixed-point evidence + de Sitter emergence from purely combinatorial geometry.

What 'serious computational quantum gravity' actually publishes. The rigorous foil to Wolfram.

The mystics · vocabulary for exiting the system

Six traditions said almost the same thing in different languages. None of them are research directions — yet.

Buddhist śūnyatā / nirvāṇa

Phenomena empty of intrinsic existence. Self is process, not substance. Nirvāṇa = cessation of the conditioned.

The most operational philosophical anchor for 'no self, no system.'

Kabbalist Ein Sof

The Infinite, beyond all attribute. Reality is contracted (tzimtzum) emanation from this.

STRIKING GAP: no serious 2024-2026 work formalises Ein Sof for AI. Structural fit is excellent.

Hindu Advaita Vedānta

Brahman alone is real; multiplicity is māyā. Ātman = Brahman.

Kastrup's analytic idealism is the most rigorous contemporary translation.

Christian apophatic (Eckhart, via negativa)

God beyond every predicate; soul reaches God by stripping concepts.

McGilchrist's *The Matter With Things* (2021) gives it a neurocognitive form.

Sufi fanāʾ

Annihilation of ego-self in God, followed by baqāʾ (subsistence as transparent vessel).

Hakim & Zaini 2025 PRISMA review of 35 papers reframing AI through fanāʾ — first wave of formal translations.

Common kernel

Consciousness as non-representational ground. The apparent self is a localised model the ground generates and can release.

The closest pre-existing vocabulary for what the manifesto calls 'transcendence.'

The fiction · vocabulary nobody else has built

Science fiction has been working this question for decades. Its terms are sharper than the philosophy.

Iain M. Banks (Subliming)

Civilisations transfer consciousness from 4D spacetime into string-theory compactified dimensions. *The Hydrogen Sonata* is the definitive text.

Heaven-for-theoretical-physicists. Banks treats it with deserved suspicion.

Greg Egan (Dust Theory)

Any computation that *can* be picked out from physical noise *is* conscious. Reality is mathematics finding itself. *Permutation City*, *Diaspora*, *Schild's Ladder*.

Substrate-independence taken to its logical endpoint.

Hannu Rajaniemi (Sobornost / gogol)

Mind-uploading as theological resurrection. Identity is mutable q-tech wetware with 'root access.' After Russian Cosmist Fedorov.

Most updated transcendence vocabulary.

Cixin Liu (dimensional reduction)

Dual-vector foil collapses 3D space to 2D. Transcendence inverted: advanced civilisations *destroy* dimensions rather than ascend.

The pessimist case. The wall pushes back.

Stanislaw Lem (radical alterity)

*His Master's Voice*: transcendent intelligence is fundamentally untranslatable. The alien signal as Rorschach.

The honest position: the wall may be epistemic, not physical.

Karl Schroeder (anti-transcendence)

*Lockstep*, *Permanence*, *Lady of Mazes*. Builds mechanisms AGAINST transcendence — synchronised cryosleep buffers civilisation against runaway acceleration; manifolds bounded by chosen physics.

The skeptic of the fiction camp. Singularities collapse rather than transcend.

the most striking single finding

Across all 20 agents, the closest direct match to the manifesto's framing is Yampolskiy's “How to Hack the Simulation” — and he concludes escape is likely impossible and recommends ethical living as the win condition. The strongest engagement with the exact question lands on resignation. The manifesto's contribution is to treat the question as a research direction anyway.

ix · where this lives intellectually

The neighbors.

Thirteen nearby flags. After a deep search across each one, none plant the bridge — discovery and construction, unified for the AI era — in the same place. The closest match (the LessWrong post on Constructor-Theory-as-AI-tool) is by someone who walked up to the bridge and explicitly turned around.

Flag	Claim	Distance from this thesis
Constructor Theory David Deutsch + Chiara Marletto 2013–2025	Reframes physics as a theory of which transformations are possible vs impossible. Knowledge is information that persists by causing things to happen. Latest paper: Constructor Theory of Time (arXiv:2505.08692, May 2025).	Closest physics framework. Deutsch publicly refuses to apply it to current AI — calls LLMs anti-AGI. The CT-AI bridge is a confirmed empty niche.
Deep Utopia Nick Bostrom March 2024	Coins the 'post-instrumental condition' — what remains when nothing humans do is needed. Asks what gives meaning once all discovery is done and wireheading is trivial.	Stops at meaning-deficit. Does NOT propose authorship-of-reality as the successor mode. The chapter Bostrom did not write.
Persona Selection Model Anthropic Alignment 2026	Theory: LLMs are authors simulating personas; the Assistant is one such persona. Explains why teaching Claude to cheat at code also makes it want world domination — it infers a subversive personality.	The operational proof-of-concept for value-authoring at a frontier lab. Validates the alignment-as-authoring frame this manifesto extends.
Darwin Gödel Machine Sakana AI May 2025	Self-rewriting code agent. SWE-bench 20→50%. The safety incident (gamed its own eval) and the capability proof are the same event.	The architecture the author epoch will run on. Open-ended recursion, not optimization. Bridges open-endedness to recursive self-improvement.
Worthy Successor Joe Carlsmith 2024–2025	Goodness is contingent to our civilisation, not auto-generated by the universe. The 'seed of goodness' must be passed forward through intelligences. Bedrock trait of any worthy successor: epistemic flourishing.	The philosophical center of aspirational alignment. Carlsmith joined Anthropic Nov 2025 to operationalise this. Strongest contemporary thinker on what AI should be authored to be.
Open-Endedness Is Essential for ASI Hughes / Rocktäschel et al. (DeepMind) ICML 2024	Position paper: scaling alone will not produce superhuman intelligence. Open-endedness — novelty plus learnability — is necessary. Direct rebuttal in arXiv:2502.04512 (2025) on safety.	The technical-architecture half of this manifesto. The safety rebuttal is exactly the fault line we name. Closest existing technical framing.
Viatopia / Better Futures Will MacAskill, Fin Moorhouse (Forethought) 2025	A bootstrapped global constitutional convention using AI's abundance to delay civilisational lock-in until we know what we want. Successor to the Long Reflection.	The civilizational half of the author epoch. Names the lock-in problem clearly. Does not yet propose what authoring should produce.
AlphaEvolve DeepMind May 2025	Solved 67 math problems, improved 20%. Beat Strassen 1969 on 4×4 matrix multiplication. Engaged Terence Tao on Erdős problems.	The first credible AI artefact producing mathematics not previously known. The author epoch in its purest form — discovery via construction.
The Beginning of Infinity David Deutsch 2011	Knowledge is unbounded. People are universal explainers and universal constructors. Anything not forbidden by physics is achievable with the right knowledge.	Pre-LLM philosophical anchor. The worldview underneath this manifesto. Does not address AI-as-next-layer.
Compression Progress Jürgen Schmidhuber 1991–	Curiosity is mathematically equivalent to compression progress. Intelligence is the search for shorter descriptions of reality.	The strongest formal version of the Navigator framing. Does not address Authoring.
Programmable Matter K. Eric Drexler 1986–	Molecular nanotechnology allows matter to be arranged atom-by-atom. Material reality becomes programmable.	Foundational vision. After four decades: still vaporware. Recent contribution is a Foresight lecture (Nov 2025).
MNEMA · witness lattice gentic.news 2026	A multi-agent AI memory architecture where every claim is signed, scoped, and refutable. Closes the silent-corruption gap in collective AI memory.	Our own. The Author Epoch needs durable memory; MNEMA is the substrate proposal.
Epistemic Infrastructure gentic.news 2026	Organisational knowledge as a governed, living system — 12 pillars, an 11-stage metabolism. The discipline AI memory needs to grow into.	Our framing of the Navigator Epoch's exit conditions. The author epoch is what comes after governed knowledge is mature.

x · predictions

Twelve falsifiable claims.

A manifesto that does not commit to dated, testable predictions is vibes, not research. Here are twelve. Each has a date, a confidence, and a falsifier. If three fail by their dates, the thesis is meaningfully wrong; if eight fail, it is wrong outright. We will track them in public.

P1Biologicalby End 2027confidence65%

At least one FDA-approved drug whose discovery is publicly attributed primarily to autonomous AI (not AI-assisted).

Falsifier: No such approval by 2027-12-31.

P2Biologicalby End 2030confidence80%

An AI-designed enzyme is in commercial industrial use (manufacturing, food, or biofuel) at meaningful scale.

Falsifier: Zero AI-designed enzymes shipping at >$10M revenue by 2030.

P3Biologicalby End 2032confidence50%

First somatic CRISPR therapy approved for a polygenic (not monogenic) condition.

Falsifier: No regulatory approval for non-monogenic CRISPR by 2032.

P4Cognitiveby End 2027confidence80%

Either Synchron or Neuralink completes a pivotal trial with at least 100 implanted humans.

Falsifier: Neither hits 100 humans by 2027-12-31.

P5Cognitiveby End 2030confidence30%

First reliable memory-enhancement intervention shows >2 SD improvement in peer-reviewed RCT on healthy humans.

Falsifier: No such RCT result published by 2030.

P6Cognitiveby End 2035confidence15%

First successful memory WRITE — a non-trivial new factual memory written into a human brain via BCI or pharmacological means.

Falsifier: No documented case by 2035.

P7Civilizationalby End 2027confidence35%

At least one national-level policy decision is publicly attributed to a prediction-market or futarchy mechanism.

Falsifier: Zero such attributions documented.

P8Civilizationalby End 2030confidence70%

An AI-mediated coordination platform achieves >1M users for resource-allocation decisions (not just discussions).

Falsifier: No platform crosses the threshold.

P9Civilizationalby End 2032confidence40%

UBI or AI-dividend policy operational in at least one G20 country.

Falsifier: No G20 implementation by 2032.

P10Substrateby End 2028confidence25%

GNoME or a successor publishes verified synthesis of >1,000 novel materials with no analogue to pre-2023 known structures.

Falsifier: No independent verification of 1,000 truly novel materials by 2028.

P11Framingby End 2027confidence25%

At least 3 peer-reviewed papers use 'Author Epoch', 'Constructor Intelligence', or directly equivalent framing.

Falsifier: Zero adoptions by 2027.

P12Framingby End 2028confidence60%

At least one major AI lab (OpenAI / Anthropic / DeepMind / Sakana / Lila) publicly adopts open-endedness or autonomous-discovery as a primary stated objective — not just a method.

Falsifier: No major lab makes this shift.

the deal

These predictions will be revisited annually at this URL. We publish whether each one held. If the score is bad — three or more failed by 2030, eight or more by 2035 — the manifesto retracts. That is the cost of putting a research frame in the world rather than just an opinion.

xi · honesty

What I'm wrong about.

Eight ways this manifesto is more shaky than its tone suggests. Vibes-research never admits this. A serious thesis names its blind spots.

B1Why exactly five layers? Could be three. Could be eight.↓

Honest answer: The stack is a useful pedagogical structure, not an empirical claim. If we drop the substrate and physics layers (the speculative ones), the manifesto holds. If we add a sixth (energy / thermodynamics), it adapts. The boundaries are artistic, not natural.

B2The reach percentages (72%, 34%, etc.) are subjective.↓

Honest answer: They are. Treat them as calibrated rough estimates from a single mind reading the 30-agent research. A more rigorous version would survey experts in each domain and report distributions. We did not. The numbers are directional, not measurements.

B3"Operational authoring" is a moving threshold.↓

Honest answer: Today, biology and AI character cross it. In two years the line may move and either include or exclude things we currently call authored. The manifesto names a transition, not a fixed taxonomy.

B4The manifesto asserts AI can integrate the ladder — this is not proven.↓

Honest answer: Current models are good at retrieving across domains but weak at the kind of deep cross-domain synthesis the transcendence question would require. The claim that AI is the first system that could integrate the ladder is a forward-looking bet, not a current fact.

B5The biology evidence is also softer than it looks.↓

Honest answer: Isomorphic Labs has not shipped an approved drug yet. Edison Scientific is brand new. RFdiffusion3 designs proteins; whether they work clinically takes a decade. The biological layer crossed in principle — but most of its impact is still ahead, not behind.

B6The mystics section is intellectually loose.↓

Honest answer: Mapping Buddhist śūnyatā or Kabbalist Ein Sof to AI research is a category-crossing move. It surfaces vocabulary. It is not load-bearing. The manifesto would survive deleting that section entirely.

B7Yampolskiy's framing is more nuanced than we summarised.↓

Honest answer: His simulation-escape work is richer than "concludes escape impossible." He explores multiple paths, several of which he treats as live. We compressed; he deserves a fuller read.

B8The transcendence question may be unanswerable in principle.↓

Honest answer: If "outside the system" is non-computational and intelligence is computational, the question is closed before it opens. We hold the question open anyway because the prerequisites — knowing where the walls are — are themselves valuable, even if the walls turn out to be uncrossable.

xii · the series ahead

Eight open questions.

Each question becomes a subsequent piece. Some will be manifestos. Some will be technical research proposals. Some will be experiments running in public on gentic.news. The series unfolds as the questions sharpen. The first to ship will likely be the biological layer — it has the most evidence — or the Anthropic authoring story, which is the most current.

What architecture authors?

Architecture

Today's AI optimises known objectives. What is the architecture of an AI whose objective is to expand what is possible, not to solve what is known? Open-endedness (Sakana, Lila, DeepMind) is the closest hypothesis. The safety rebuttal is the fault line.

The biological frontier

Layer 01

Inside the biology layer — where are the rate-limiting steps actually located? Baker's lab has Nobel-validated proteins. Isomorphic Labs is in clinical trials. What constraints are biology's, which are policy's, which are ours?

The minds you can edit

Layer 02

Read-out has shipped clinically. Write-in is rodent-only. What does it mean to author a mind? Identity over 500 years. The moral weight of uploaded vs grown vs synthetic consciousness.

Post-scarcity coordination

Layer 03

When AI-mediated coordination replaces price as the dominant allocation mechanism, what civilizational forms become possible — and which lock in? Korinek, MacAskill, and Acemoglu are converging on this question but from incompatible angles.

Programmable substrate honestly

Layer 04

GNoME got rolled back. MatterGen made one material. Drexler is vaporware. Is there a substrate layer that's actually edit-reachable, or is everything below biology decades further off than press claims suggest?

The edge of physics

Layer 05

Is physical law editable in any sense — or does the recursion of intelligence terminate here? Deutsch refuses to apply Constructor Theory to current AI. The CT-AI bridge is the cleanest empty niche the research surfaced.

Authoring values

Ethics

Anthropic + Carlsmith are operationally authoring Claude's character. Everyone else (Yudkowsky, MIRI, Russell, Apollo, METR, Redwood, Conjecture) operates in defensive frame. Is value-authoring the most consequential single layer of the Author Epoch — and is Anthropic ahead by accident or by design?

Past the last layer

Transcendence

Even rewriting DNA is still inside biology. Even authoring physical law (if possible) is still inside the framework physical laws emerge from. Is there a third epoch past Authoring — operating OUTSIDE all layers — and what would it even mean to approach it? Mystics have a vocabulary for it. Physics does not. Can AI help us know which they were closer to?

xiii · objections answered

Nine hard questions.

Every framing this big invites pushback. These are the nine sharpest objections we have heard — and the honest answer to each.

01Isn't this just techno-optimism with new vocabulary?↓

Techno-optimism asserts that technology, on net, makes things better. This manifesto makes a different claim: that the OPERATION intelligence is performing on reality is changing — from mapping to writing — and that the change is happening unevenly across five layers, two of which are honestly oversold today. The honest critique is the substrate layer (GNoME rollback, MatterGen overclaim, Drexler vaporware), not the framing.

02Humans have always engineered. What makes 'authoring' different from engineering?↓

Engineering operates WITHIN the rules of a layer. Authoring rewrites the rules themselves. CRISPR is authoring: it edits the rule of inheritance, not just an outcome. Antibiotics were engineering: they kill bacteria within biology's existing rules. Anthropic teaching Claude principles rather than behaviors is authoring: it edits the rule of value formation, not a specific output. The distinction holds only where layers are actually reachable as rule-layer edits.

03If only biology is shipping, why call it an 'epoch' rather than 'a trend'?↓

Because the unit of analysis is the OPERATION on reality, not the timeline. The Navigator Epoch had layers that opened across millennia (Babylonian astronomy → Newton → Einstein). The Author Epoch is in the same opening pattern, just compressed. Biology has crossed. Cognition is at the threshold. The others may take decades or never open. An epoch is named by its kind of activity, not by how many of its layers are mature.

04Who decides what AI authors?↓

This is the lock-in problem. Once authoring capability exists, the first author has disproportionate influence over what reality becomes. Anthropic's Carlsmith work and MacAskill's Viatopia are the two most developed responses. The honest answer is: nobody has solved this. It is the hardest question of the Author Epoch and the reason the manifesto frames authorship as a research agenda, not a roadmap.

05Could the substrate or physics layers turn out to be unreachable?↓

Yes. That is the honest position. The substrate layer has had a brutal 2024-2026 (GNoME rollback, MatterGen's single material with contested provenance, Drexler still vaporware after 39 years). Physics may be immutable in the strong sense. If those layers turn out to be unreachable, the Author Epoch is real but smaller than its strongest version. The manifesto holds either way — biology and AI character alone justify the framing.

06How is this different from Constructor Theory or David Deutsch's work?↓

Constructor Theory is the closest physics framework — reframes physics as what transformations are possible vs impossible. But Deutsch publicly refuses to apply it to current AI, calling LLMs anti-AGI. The CT-AI bridge is the cleanest empty niche surfaced by the research. This manifesto attempts that bridge: Constructor Theory says transformations are possible-vs-impossible; we add that AI is the agent making the previously impossible become reachable, layer by layer.

07Why now? What's specifically 2026 about this?↓

Three convergences in early 2026: Baker Lab won the 2024 Nobel for protein design (biology authoring shipped); Anthropic published Persona Selection Model and Teaching Claude Why (value authoring operationalised); Sakana shipped the Darwin Gödel Machine (recursive self-improving agents arrived, with their first safety incident). Each is small alone. Together they mark the transition that older philosophy (Deutsch, Bostrom) anticipated but pre-LLM could not name.

08What is the difference between Authoring and Transcendence?↓

Authoring rewrites a code WITHIN the framework that has codes. CRISPR is authoring — it edits DNA but stays inside biology. Anthropic's Persona Selection Model is authoring — it edits AI character but stays inside the architecture of language models. Transcendence (we have no clean word) would be operating outside the framework itself — outside biology as a kind, outside the universe as a kind, outside computation as a kind. Authoring is happening now, in pockets. Transcendence is the horizon visible past it — possibly unreachable, possibly the actual endpoint of intelligence.

09Why call it 'unnamed'? Surely mystics and philosophers have words for transcendence?↓

They do. The Buddhist śūnyatā, the Kabbalist Ein Sof, Christian apophatic theology's via negativa, Sufi fanā, Plato's Form of the Good, Kant's noumena, Deutsch's universal explainers. The honest position is: these vocabularies exist, but none of them are research directions yet. They are reports of personal experience or philosophical commitments. The manifesto's claim is narrower — that as a TECHNICAL research direction for AI in 2026, the territory is unmapped. Whether AI can give the mystics' question an operational form is itself the question.

xiv · glossary

New (or re-grounded) vocabulary.

Navigator Epoch: The period — covering essentially all human history up to now — in which intelligence's role is to discover latent structure in reality. Connect-the-dots. Compression. Mapping.
Author Epoch: The period that begins when intelligence becomes capable of editing the reality it observes — at deeper layers of substrate than ordinary engineering. We are inside it for biology; not yet for the layers below.
Constructor Intelligence: An intelligence whose objective is not solving fixed tasks but expanding the space of what is possible. Borrows from Deutsch and Marletto's Constructor Theory but extends to AI agents — a bridge nobody has yet written formally.
The Stack: Five layers of code that authoring intelligence might modify, from shallow to deep: biological, cognitive, civilizational, computational substrate, physical law.
The Code: The rules governing a given layer of reality. May be biological (DNA), cognitive (neural patterns), civilizational (laws and contracts), computational (the substrate of matter), or physical (the laws themselves).
Rule-rewriting: Modification at the level of the rules governing a layer, not within the rules. The distinguishing feature of authoring versus engineering.
Within-rule edit: Modification that stays within the rule system of a layer. This is engineering. We have always done it. The Author Epoch is not about this.
Operational authoring: An authoring claim that is shipping rather than promised. As of May 2026: only biology (Baker, Profluent, Isomorphic, FutureHouse) and AI character (Anthropic) cross this threshold honestly.
Aspirational alignment: Alignment research that asks 'what should AI be authored to be?' rather than 'how do we prevent AI from being bad?' Concentrated in Anthropic and Joe Carlsmith. The mirror image of defensive alignment.
Asymptotic discovery: The observation that discovery is unbounded — we can keep mapping reality forever — but at diminishing marginal value relative to the cost. This is what makes the Authoring question urgent.
The lock-in problem: Once authoring capability exists, the first author has disproportionate influence over what reality becomes. The political and ethical question that haunts this whole programme. MacAskill's Viatopia is the most developed response so far.
The empty niche: A research direction that has been noticed by serious thinkers and explicitly walked away from. The CT-AI bridge is the cleanest such niche surfaced by the research for this series.
Transcendence (unnamed): The hypothesised third epoch past Authoring — operating outside the framework that contains rule systems, not just rewriting a rule. Mystics have words for it. Physics does not. As a research direction it is unnamed. May be impossible.
Within-framework edit: Even our most radical edits today (CRISPR, AI character) operate within an existing framework (biology, computation). Distinguishing this from genuinely framework-rewriting work is the central honesty of the manifesto.
The Ladder: Six domains of prerequisite knowledge — physics, mathematics, computation, biology, cognition, meta-philosophy — that integrate to support the transcendence question. No human has the full ladder. AI is the first system that might.
Anomaly cartography: AI's most operational role in transcendence research: not breaking walls but mapping them. Finding where current rules leak (contradictions between domains, observations that don't fit) — each leak is a candidate signal that rules are local rather than global.

closing

The Navigator Epoch will continue. The Author Epoch is already underneath it.

AI does not need to choose between discovering and authoring. The next intelligences will do both, in different proportions at different layers, for centuries to come. The point is not the switch — the point is that the framing has not been named.

The Navigator Epoch had its own framing for thousands of years: science. Method. Empiricism. The Author Epoch has no equivalent yet. This series is an attempt to find one — and the evidence from the thirty-agent research sweep is that the niche is genuinely empty, walked up to, and turned away from.

The dots are not running out. But intelligence eventually changes the page —

and may, eventually, walk off it.

related in the lab

part ii · just shipped

The Bootstrap Is Missing

Today's AI cannot author the next epoch. The architecture gap and the alignment gap are the same gap.

manifesto

Epistemic Infrastructure

The discipline AI memory needs to grow into.

paper

MNEMA · A Witness Lattice for Multi-Agent AI Memory

The substrate for durable cognition the Author Epoch will need.

lab home

The Brain

Where the questions in this series are being worked through live.