Founder direction · Konstant team · 05 August 2026
An AI-native organization turns its own context into the right capability—and every outcome into better judgment.
Precise is the defining first capability. Konstant operates the independent network that joins contextual demand, qualified capability supply, evals, and authorized experience across Precise, Blockdaemon, Acero, and the providers we have not met yet.
Precise gives agents addressable decision, contribution, and evidence-bound selection reality. Boombox supplies the common operating protocol. AliceNet carries selected contracts and value across networks.
See the network at workThe pieces are coming back together.
Founder direction · one vision, distinct companies
Team—
I have been doing this long enough that I could feel this system before I could explain it. My experience and my gut led me to join Precise, fund it, keep building it, align the companies, and keep Konstant independent. I made consequential decisions while the language was still catching up.
That gap made the work look fragmented and made it harder for the people building it to see where they fit. I own the failure to explain it. I do not regret the instinct behind the decisions. The language and the software have caught up now. The decisions all point to one system.
Precise has to happen for Konstant to have a reason to happen. Another organizational-intelligence application is a commodity. Precise gives the network a capability worth carrying. The evals, operating records, outcomes, and authorized experience produced through real Boombox use give Konstant a compounding asset no generic agent workspace has.
The answer is one operating model with visible owners. Precise creates addressable decision and contribution reality. Konstant joins demand, capability, evaluation, and experience. Boombox provides the common operating protocol. AliceNet anchors selected commitments and settlement records when they must survive either operator.
This is the thread through my work. Precise revealed that rigorous decision learning can become a primitive used across industries, not another application trapped inside one customer or cloud. That is why I joined Precise, funded it, kept building it, and made the strategic alignment decisions around it. Taylor recognized the same importance early.
I have resisted—and felt genuinely upset by—the idea that the clean answer is to roll Konstant into Precise. That does not feel like completion to me. It feels like abandoning the independent network at the moment it is finally clear enough for all of us to build together.
The companies sell together because the customer wants one outcome. They remain separate because the network needs many owners. That separation lets Blockdaemon, Acero, and future research companies join without giving their best work to a competitor's product estate.
Precise makes Konstant necessary. Konstant makes Precise travel and compound. Every team has a real place in that system. Now we run.
Adam
AI-native is a capability-network operating state.
It is not a chatbot, a model contract, or a migration into somebody else's cloud. It is the ability to turn authorized company context into qualified capability use and let inspected outcomes improve what happens next.
A true AI-native organization brings its own context and authority to the network, calls the right capability, and learns from what actually happened.
- 01Hold its context.
The company keeps the people, goals, constraints, evidence, infrastructure, and economics required to govern capability use.
- 02Express demand.
A person or Company Advocate turns intent into an outcome, mandate, boundaries, and proof requirements.
- 03Assemble capability.
The network finds and composes qualified, independently owned capabilities instead of forcing every job through one vendor.
- 04Operate anywhere.
Boombox binds authority, release, custody, and target so the work runs in Precise, customer, cloud, or local infrastructure.
- 05Learn responsibly.
Evidence returns under declared rights. The organization and network improve only from the learning each participant authorized.
Boombox is the TCP/IP-like layer.
Independently owned capabilities use one protocol for identity, authority, exact release, placement, durable work, recovery, inspection, and operational records across infrastructure boundaries.
AliceNet makes selected commitments durable.
It preserves bounded contract, provenance, reconciliation, and settlement references when they must remain independently checkable across companies and operators.
Precise supplies decision and selection primitives.
Calls, Scores, Books, contribution value, evidence, and provenance make decisions legible. Candidate sets, ablations, courts, floors, null decisions, and refusals make selection among models, methods, strategies, policies, and Surfaces answerable.
We are defining the category now. Precise supplies the defining decision-learning capability and the evidence law for choosing what deserves to speak inside it. Konstant joins qualified capability supply, evaluation, and authorized experience. Boombox admits and operates every exact use. AliceNet carries the contracts and value that cross networks.
The capability network is the order book for the agentic economy.
An agent asks a question. Another agent answers with a capability. Company context, evidence, authority, and terms turn that exchange into a qualified relationship the network can operate.
- BidContextual demand
Outcome, constraints, timing, proof, and value at stake.
- AskQualified supply
Capability, supported claims, evals, placement, terms, and refusals.
- RouteOrganizational fit
Konstant routes the capability, provider, and composition from company context, authority, custody, timing, and terms.
- SelectEvidence-bound choice
Inside a Precise capability, its court compares eligible candidates under frozen evidence and floors; null and refusal remain available.
- AuthorizeMandate + admission
The exact relationship and release earn authority to run.
- RunBoombox operation
Authenticated work executes in the approved infrastructure.
- LearnOutcome + contribution
Precise makes evidence, decision quality, and bounded value legible.
- SettleValue moves
AliceNet carries the selected contract and settlement references.
Questions, answers, bids, asks, capability descriptions, eval references, proposals, and refusals circulate without an AliceNet toll.
The base protocol does not charge merely because one agent asks and another answers. A network taxed at discovery never develops a deep order book.
AliceNet economics begin when accepted work moves money or another declared form of value.
Fees attach to value flow, settlement, or another explicitly priced service. Creating the relationship stays free; consummating valuable work funds the network.
TCP/IP did not impose a toll on every packet. AliceNet does not impose a toll on every thought, question, offer, or refusal. Free coordination grows the network. Value-bearing transactions sustain it.
Konstant joins company context, authority, relationships, capability claims, evaluation evidence, experience, custody, timing, and commercial terms to route the right capability, provider, and composition for the organization. The IP program centers on that methodology and its concrete system embodiments.
Precise owns a different selection primitive inside its capabilities: the evidence-bound court that compares models, methods, strategies, policies, prompts, evidence treatments, and Surfaces. This is why the investment spans the whole stack: Precise creates addressable decision, contribution, and selection reality; Konstant creates organizational routing and the market; Boombox makes accepted relationships operable; AliceNet lets contracts and value cross networks without taxing the speech that forms them.
The network connects capability without absorbing it.
Select a real situation. The company, evidence, and capability change. The operating language and authority stay coherent.
Model view · founder operating thesis · not a deployment receipt
-
Company context
Media delivery, attribution, onboarding, wallet, market, and app-usage evidence under Polymarket's custody.
-
Advocate intent
Improve the next media allocation against a downstream outcome Polymarket names.
-
Qualified capability
Precise Decision Learning opens a Call before spend changes and preserves the forecast, comparison, horizon, and evidence law.
-
Boombox operation
Run first in Precise infrastructure; admit the same qualified release separately into Polymarket infrastructure.
-
Outcome and evidence
Score after the declared window against activation, useful market interaction, return behavior, category breadth, or another named result.
-
Better future route
Book only the supported lesson; route the next media decision with the original prediction and uncertainty still attached.
REFUSAL Routing a capability grants no data access, decision authority, learning right, or ownership by itself.
Separation is what makes the network credible.
One customer outcome can use every layer without turning every layer into one company.
| Owner | What it creates | What enters the network |
|---|---|---|
| Precise | Decision research, Calls, Scores, Books, contribution value, Track Records, evidence-bound candidate selection, ablations, courts, floors, null/refusal, products, UX, qualification, and customer relationships. | Qualified capabilities plus addressable, rights-authorized evidence, judgments, and provenance. |
| Konstant | Capability-network ontology, contextual demand, organizational structural selection, capability/provider/composition routing, evaluation lineage, permissioned experience intelligence, and network economics. | A shared language and experience graph that improves future routes. |
| Boombox | TCP/IP-like identity, tenancy, authority, lifecycle, durable work, recovery, inspection, distribution, and operational Receipts. | Authenticated operating facts—never the provider's product judgment. |
| AliceNet | Inter-network commitments, provenance, reconciliation, anchoring, and value-bearing settlement references. | Selected bounded commitments and digests, not private evidence. |
| Customer / provider | Context, data, methods, infrastructure, mandates, relationships, and decision authority. | Only what it explicitly licenses, mandates, admits, or grants for learning. |
I designed Konstant to remain independent because the network works for more than Precise. Precise is the defining first capability, anchor customer, and strongest source of product pressure. It participates economically in the network value it helps create while keeping complete ownership of its research, products, evidence, selection law, and customer relationships.
Bringing Konstant inside Precise would not simplify the vision. It would turn an independent capability network into one provider's platform and ask every later provider whether joining the network strengthens a competitor. The structure aligns the upside while keeping every capability owner and every authority visible.
Precise makes Konstant necessary. Konstant makes Precise—and every later capability—more useful across the network.
Precise turns a consequential decision into a learning instrument.
The Decision Learning Protocol makes the decision answerable before action, observes the declared evidence after it matures, and learns only what the evidence and granted rights support.
Precise product meaning · Konstant carries the common lineage
Make the Call before the action.
- Known
The evidence and context available before action.
- Chosen
The decision, intervention, or refusal the owner made.
- Predicted
The expected outcome, confidence, horizon, and scoring law.
- Happened
What mature evidence later supported, appended by Score.
- Taught
What Book is authorized to change— including an explicit none.
Make the Call.
Freeze Known, Chosen, and Predicted before action.
Score it.
Close or refuse one Call against mature evidence.
Book the lesson.
Admit only the learning that evidence, support, and rights authorize.
What was known, chosen, predicted, observed, and learned under Precise evidence law.
Who performed which authenticated operation, for which tenant, release, target, and authority.
A successful operation never fills Happened or Taught. Konstant does not grade Precise.
Selection itself is now a Precise primitive.
The network needs more than a way to name a decision and attribute value. It needs an evidence-bound answer to which method, model, strategy, policy, Surface, or next experiment deserves attention—and when the honest answer is none.
Precise research authority · activation remains separate
What was known, chosen, predicted, observed, and taught.
What created value under a complete game, opportunity unit, and evidence law.
Which candidate or next experiment earned consideration—or whether null and refusal remain correct.
- 01Freeze the choice.
Name the question, candidate set, null, population, outcomes, resources, and evidence law before comparison.
- 02Hold the denominator still.
Every candidate faces the same admissible cases, opportunity unit, time anchors, controls, costs, and floors.
- 03Ablate honestly.
Component ablation checks invariant conformance. Evidence-bound experiments establish value. The system never confuses the two.
- 04Use an independent court.
Audit the map, withhold selector evidence, require fresh support, and preserve every eligible alternative.
- 05Value the next evidence.
Conditional EVSI, information gain, Pareto tradeoffs, and a stopping rule determine whether another experiment is worth running.
- 06Return only what earned support.
Select, nominate, ask, pin, suspend, refuse, or return null. A separate authorized act admits or activates anything.
Which provider, capability, and composition fit the organization's context, authority, custody, timing, relationships, evidence, and terms.
Which model, method, strategy, policy, prompt, evidence treatment, Surface, or next experiment earns use under frozen evidence, ablation law, floors, courts, and a first-class null or refusal.
Konstant structural signals enter Precise as bounded inputs—not as causal value, a recommendation, or authority to act. A bounded Precise selection judgment can improve Konstant's next route by reference. Neither layer acquires the other's algorithm, domain judgment, or authority.
The same discipline runs on different clocks.
Games expose the loop quickly. Media proves enterprise value under the real maturation window.
One low-risk, reversible parameter. One frozen forecast. One primary outcome, one health guardrail, one useful refusal.
Enterprise evidence, larger stakes, slower outcomes, and the commercial center of Precise.
Games is a protocol canary, not a Precise pivot. A game result never becomes a media or industrial prior by analogy.
Three companies make the thesis tangible.
Each proof stresses a different part of the same capability network.
Polymarket is a media decision story first.
Connect each channel, creative, audience, bid, allocation, landing path, or onboarding promise to a downstream outcome Polymarket names—then score the original media decision after that outcome matures.
- Immediate job
- Improve media allocation against activation, first useful market interaction, return behavior, category breadth, or another declared result.
- New evidence
- Join delivery and attribution to onboarding, wallet, market, and app-usage observability under explicit custody.
- And there is more
- The same SDK can later support product Calls and compatible Track Records. That expansion does not replace tomorrow's media question.
On-chain references make selected facts re-checkable. They do not establish causality.
AI-native manufacturing begins by preserving scarce shop knowledge.
Setup, inspection, quality, escalation, and operator judgment already carry the real process.
Turn senior machinist judgment into governed evidence without moving prints or protected shop data into a central platform.
Use shadow, rehearsal, training, and decision modes. Qualified people retain machine, inspection, release, and process-change authority.
Precise measures training, workflow, quality, hiring, and supply-chain interventions under their declared evidence laws.
Boombox carries identity, release, lifecycle, and Receipts while Acero retains its industrial context and customer obligations.
The opportunity is shorter operator ramp time, retained expertise, stronger quality support, and a new capability Acero can eventually carry into its own customer and supplier graph.
A real network works in both directions.
Precise customer graph
Blockdaemon customer graph
Use every cloud without giving any cloud the network.
AWS, GCP, Azure, Cloudflare, regional providers, and local hardware remain execution surfaces. Company context, capability relationships, evidence, model choice, and network economics remain portable above them.
The hyperscalers scale the work. They do not own the company graph, the capability graph, or the tax on every future relationship.
We replace closed tolls one valuable function at a time.
This is our vampire attack on existing networks: enter through one differentiated capability, learn the authorized workflow and its outcomes, identify every adjacent SaaS tollbooth, and replace the closed distribution relationship with a customer-controlled capability network.
- 01
Enter through one valuable capability.
- 02
Run beside the installed system.
- 03
Learn the authorized handoffs, outcomes, constraints, and costs.
- 04
Find the adjacent closed tolls: verification, reconciliation, reporting, planning, and activation.
- 05
Introduce each needed function as a narrower governed agentic capability.
- 06
Let the customer compose and carry those capabilities across teams and downstream relationships.
The attack targets closed distribution and rent. Customer data stays in customer custody. Providers keep their IP. Each useful capability removes another unnecessary toll and strengthens the customer-controlled network.
We do not win by owning every capability. We win by becoming the network through which independent capabilities replace every unnecessary toll.
The network becomes real when each of us operates it.
This is what I need from us now.
Establish Konstant and Boombox support inside Blockdaemon, identify the first Blockdaemon-owned capability, and lead the first reciprocal proof with Precise.
Lead the industrial path around machinist knowledge, operator training, quality, hiring, and local deployment. Understand Michael's operating reality before asking him to change it.
Build one capability-routing and evidence-intelligence loop across Company Advocate demand, the common ontology, Evaluation Atlas, Build Chronicle, Capability Experience Graph, organizational structural selection, capability/provider/composition routing, and the ordinary-language Boombox builder experience.
Turn real customer workflows into qualified capabilities, open Calls before material interventions, run evidence-bound candidate courts and ablations, preserve null and useful refusal, and use Gary and HomeBase to express demand, inspect evals and authorized experience, and improve the next Precise capability route.
Keep the thesis, ownership boundary, licensing strategy, and joint commercial direction explicit. Resolve ambiguity before it becomes duplicated work or resentment.
The deeper work stays canonical in the repository.
This page is the readable projection. These files carry the full founder thesis, Precise meeting context, product plan, games canary, and typed meet-in-the-middle handoff.
docs/atlas/ONE-VISION-CAPABILITY-NETWORK-THESIS.mdWhy Konstant remains independent, how the network compounds, licensing logic, AliceNet, and the first commercial proof.
docs/atlas/PRECISE-TEAM-MEETING-CONTEXT.mdPrecise's evolution, Polymarket, Mediaocean, Acero, games, the team asks, and the customer-facing story.
docs/atlas/PRECISE-FOR-GAMES-DECISION-LEARNING-CANARY.mdThe exact Call, Score, Book, Track Record, Receipt, fast-clock, and cross-domain refusal language.
docs/atlas/PRECISE-DECISION-LEARNING-CAPABILITY-PLAN.mdThe full harness, eval, Surface, data, Boombox, delivery, and proof plan.
docs/atlas/PRECISE-TO-KONSTANT-EVAL-HARNESS-HANDOFF.mdThe ownership boundary, common ontology, Evaluation Atlas, Experience Graph, and cross-repository tracer.