Own Your Intelligence
A Field Guide to the Distributed AI Economy
How AI capability becomes owned infrastructure, and what becomes tradeable when it does.
Travis Oliphant, creator of NumPy and SciPy, co-founder of Anaconda, and CEO of OpenTeams, shares his vision for what's next. Learn more at OpenTeams
00 — The takeaway
Bottom line up front
Own the stack instead of renting it, so your context becomes a governed, versioned asset. An accountability plane (Guards, Gates, Tracks) checks every layer, not just the AI. And four kinds of artifact trade across independent Hubs while the evidence stays home.
The stack: accountability at every layer
The exchange: Tracks stay home
what moves: ops · cogs · frames · guards · what stays: tracks
01 — The missing returns
Everyone is buying intelligence. Few are keeping it.
56%
of CEOs report zero financial impact from AI
Source
PwC, 29th Global CEO Survey, January 2026 (n=4,454 CEOs, 95 countries). Registry entry · as of 2026-07.
12%
of CEOs achieve both cost and revenue benefits from AI
Source
PwC, 29th Global CEO Survey, January 2026 (n=4,454 CEOs, 95 countries). Registry entry · as of 2026-07.
Corporate AI investment reached $581.7 billion in 2025, up 130% in a year, by Stanford HAI's count. Yet in PwC's survey of 4,454 chief executives, most report no financial impact from it at all.
The field guide's reading of that gap: it is not only a model-quality problem. It is an ownership, context, integration, and accountability problem. An architecture problem. Better models alone will not close it.
Evidence current as of August 2026. Sources are listed with every figure. Registry
02 — Rented intelligence
The same spend. Two different balance sheets.
Consuming AI through systems you cannot inspect, reproduce, audit, or own has a compounding cost: every prompt that re-explains your organization teaches a system somebody else keeps.
The same organizational context either evaporates as an expense inside a vendor's system, or accretes as a versioned asset the organization keeps.
Rented
- prompt
- policy
- style guide
- process doc
- correction
context re-explained, retained on the vendor’s side of the wall — expensed
Owned
- prompt
- policy
- style guide
- process doc
- correction
context captured, versioned, governed — capitalized
The organization is a tenant, and tenants do not get the keys to the building's mechanical room.
03 — Context is capital
Your organization's context is an asset. Today it evaporates.
The rules, terminology, style, norms, and process knowledge that turn generic capability into your work are scattered across people, prompts, wikis, policies, and chat history — re-explained thousands of times a day, retained nowhere. The paper's own line, new in Revision 9 (§4.3): "Every prompt that re-explains the organization is capital spent as an expense."
Organizational knowledge scattered across wikis, prompts, policies and chat becomes a single scoped, versioned, governed Frame.
- wiki page
- prompt
- style guide
- Slack thread
- policy PDF
- onboarding doc
- process checklist
- brand rules
- past correction
company-context.frame
- sources:
- 9
- owner:
- knowledge-ops
- scope:
- organization
- version:
- v1.0.0
governed
Captured instead as a governed, versioned artifact, context stops being overhead and starts compounding. That artifact has a name: the Frame. It is the first piece of the architecture.
04 — The architecture
From generic capability to accountable work
Five ideas, each building on the last. No product names yet, just the shape of the machine.
1 Frames carry context
A Frame is organizational context as a versioned artifact: scoped, inheritable, composable, shareable. A company Frame cascades into department and project Frames; a regulatory Frame composes in from outside. Context flows where it should, with an audit trail.
Company Frame
"who we are, how we speak, what we never do"
Sales Department Frame
methodology, stages, definitions
Enterprise Accounts Frame
segment playbook, pricing rules
An individual’s working session
Partner Frame
installed from an external partner
Regulatory Frame
installed from an industry consortium
A cascade of Frames. A Company Frame (who we are, how we speak, what we never do) is inherited by department Frames such as Engineering and Marketing, which are inherited in turn by team and project Frames. Separately installed external Frames, such as a regulatory Frame, compose alongside the internal cascade for a given piece of work. Context flows down the organizational structure with an auditable inheritance chain.
2 Cogs perform bounded work
A Cog is an AI worker you can hold to account: a model, oriented by Frames, with declared tools and permissions. Not a bare model: an assembly. Ask "what did this Cog do, under which Frames, with what outcome?" and the question has an answer.
A Cog is an assembly of a model, Frame-founded context, declared tools, and governance parameters. It is the unit at which AI work becomes auditable.
cog: doc-collector v1.2.0
- modelpossibly specialized for the task
- contextfounded on one or more Frames
- harness · skills · tools & APIswhat it is built on and may call, and nothing else
- governancedata access, permitted actions, approvals
3 Ops package the job
Knowledge work comes in jobs, not tasks: "close the books," "run the quarterly compliance review." An Op packages a whole job: Cogs, workflow Frames, human checkpoints, integration logic, and a declared validation strategy, versioned and installable. "Notice what you cannot do with a job today: buy one."
4 Guards check. Gates decide. Tracks record.
A fluent output and a correct output look identical at a glance. Institutions that face auditors are not permitted to operate on glances. So verification is its own layer, and failure is part of the record.
Guards verify the work, Gates decide what happens next, and the Track preserves the whole sequence — failures included.
Track: evidence, appended as it happens
- guard: source-grounding — pass
- guard: sensitive-data — FAIL
- gate: revise — returned to work
- guard: sensitive-data — pass (rev 2)
- gate: proceed
The failed check stays in the record. That is the point.
5 The Intelligence Hub keeps it yours
All of it runs inside an Intelligence Hub, a governed deployment within your own perimeter, where Frames live, Cogs run, Ops execute, and the evidence accumulates as Organizational Memory. Sovereignty here is realistic, not absolutist: control the model choice, the data paths, the policies, and the evidence without manufacturing chips or training a frontier model.
Only now do the proper nouns matter. The Hub is assembled from the open-source commons: Nebari, the infrastructure stack, and Nebi, the packaging and environment layer. What exists versus what is proposed is tracked below, plainly.
Frames guide the work. Cogs perform the work. Ops orchestrate the work. Guards verify the work. Gates decide whether the work proceeds. Tracks make the work accountable.
05 — One job, end to end
Watch every abstraction become one job
The same example runs through this whole site: a quarterly compliance review, packaged as an Op. It is illustrative (no shipped product bears this name), but every term above has a seat in it.
An Op coordinates Cogs, Frames, Guards, Gates, and a human checkpoint into one accountable, installable job; a Track records everything that happened.
op: quarterly-compliance-reviewv2.3.1
- proceed
- request revision
- require human review
- escalate to expert
- stop
Track: evidence, appended as it happens
- frames: company-policy, regulatory-2026
- cog: doc-collector — invoked
- cog: policy-checker — invoked
- cog: exception-summarizer — invoked
- guard: source-grounding — pass
- human: compliance reviewer — approved
- gate: proceed
- output: findings report — delivered
The guide's own self-test: cover the figure and explain why an Op written by a stranger can be safely installed in your organization. If your answer touches a manifest, a Frame, a Guard, and a Track — you have the architecture. Read it in full
06 — The whole stack
Drawn on one page
Layer 3 — Economy §10–12
The marketplace: Ops, Cogs, Frames and Guards published, discovered, installed across sovereign Hubs. Tracks stay home.
Interface — Products around the Hub §9
Applications, consoles, and builders — a Desktop/Web Application among them — where people meet the Hub.
Layer 2 — Execution §3–5
Ops — installable, supervised jobs
Cogs — AI workers oriented by Frames
Frames — organizational context, versioned and inheritable
Layer 1 — Infrastructure §7–8
Organizational Memory — the context the Hub accumulates
Intelligence Hub — the sovereign deployment perimeter
Nebari — curated open-source stack · Nebi — packaging and installs
Compute, open-weight models, the organization's data
Accountability plane §6
cross-cutting: it applies to every layer above
- Guards: check
- Gates: decide
- Tracks: record
The architecture drawn as one stack. Layer 1, Infrastructure (Sections 7 and 8): the open-source commons assembled into an organization-controlled Intelligence Hub, with Organizational Memory accumulating under the organization’s governance. Layer 2, Execution (Sections 3 to 5): Frames carry context, Cogs perform bounded work, Ops orchestrate complete jobs. The Interface (Section 9): the products around the Hub — applications, consoles, and builders, a Desktop/Web Application among them — through which people meet the Hub. Layer 3, Economy (Sections 10 to 12): artifacts exchanged across independent Hubs. Beside the stack, the accountability plane (Section 6) is drawn as a single panel spanning the full height, with a connector from every tier (economy, interface, execution, and infrastructure) into it: Guards check, Gates decide, Tracks preserve evidence, at every layer, for every artifact and run.
Infrastructure below, execution above it, the interface where people meet the system, an economy on top — and the accountability plane cutting across every layer. The guide keeps a legible copy
07 — The economy
Four kinds of artifact. One stays home.
Ops
trade like products: installable outcomes, priced like work
Cogs
trade like specialized labor: capability you can hire in
Frames
mostly circulate free: coordination goods that gain value by adoption
Guards
provide reusable trust: verification shared like test frameworks
Tracks
stay governed inside the organization. Evidence is not for sale.
Independent, organization-owned Intelligence Hubs exchange Ops, Cogs, Frames and Guards without merging, while each Hub's evidence record stays local.
hub A · manufacturer
- ops · cogs
- frames · guards
tracks: local
hub B · hospital network
- ops · cogs
- frames · guards
tracks: local
hub C · consultancy
- ops · cogs
- frames · guards
tracks: local
- opfrom: hub-c · v1.1
- cogfrom: hub-a · v2.0
- framefrom: hub-b · v3.2
- guardopen library · v0.9
what moves: ops, cogs, frames, guards · what stays: tracks
Beside that marketplace sits a second economy: the applications, consoles, builders, and Track stores that make a Hub usable — the products around the Hub, which the guide argues would be the larger of the two, if the Linux and Kubernetes analogies hold.
The most heavily exchanged artifact in this economy will probably never be sold.
08 — Why open
The flywheel, with its brakes shown
Platforms tax exchange; protocols compound adoption. The bet is that open standards for Frames, Cogs, and Ops do for AI work what open packaging standards did for software. The guide shows the loop and the four ways it stops, because a flywheel is a forecast, not an achievement.
- more Hubs
- bigger market for publishers
- more artifacts
- more value per Hub deployment
and back to more Hubs — a reinforcing loop
balancing forces, any of which can stall it
- cold start
- few Hubs, few publishers, little to install, weak case for Hubs
- fragmentation
- competing context and packaging standards split the network
- quality collapse
- a marketplace of mediocre Ops teaches buyers to stop looking
- incumbent bundling
- “good enough” closed features soak up the demand first
A reinforcing loop: more Hubs create a bigger market for publishers, which attracts more artifacts (Ops, Cogs, Frames, and Guards), which make Hubs more valuable, which attracts more Hubs. Four brakes can stop the flywheel: the cold-start problem (no artifacts, no buyers), fragmentation (competing incompatible standards), quality collapse (a marketplace flooded with bad artifacts), and incumbent bundling (platform vendors giving away adjacent capability). The flywheel is a forecast, not an achievement.
The guide states falsifiable indicators for whether this is working: its own, not the company's. Judge it by those
What exists, what is being built, and what remains a thesis
The guide keeps three tenses distinct, and so does this site: the infrastructure layer is real and inspectable now, the execution standards are hardening, and the economy is a forecast.
Status audited: 2026-08-18
Exists today
Nebari
Exists today — open source, in transition to a modular stack
Notes & source
Open-source (BSD-3-Clause), latest release 2026.3.1 (July 2026), per the nebari.dev audit of 2026-07-25. nebari.dev describes an ecosystem in transition: "Nebari classic" is sunsetting while the successor infrastructure layer is labeled early access by the project itself. Say "production-proven" only of the classic platform; never call the successor GA. Revision 9's maturity ledger places "Nebari and NIC (open-source infrastructure)" under Exists today, and §3.1 describes a core rearchitected since June 2025 into nebari-infrastructure-core plus "more than fifteen software packs at varying levels of maturity" — the paper's own hedge, which this label keeps.
nebari.dev · 2026-08-18
Collab Desktop App
Exists today — just released, free
Notes & source
openteams.com ships this as "Collab Desktop App — Just Released · Free": publish, discover, consume, and share agentic workflows and AI capabilities. The whitepaper (Revision 9) calls this interface layer the "Desktop/Web Application" and guide pages keep that canonical wording; the name Revision 8 used for this layer is gone from Revision 9 entirely. Revision 9 also demoted the application from THE interface to one worked example of a wider economy of products around the Hub (§7, renamed "Products Around the Hub — and the Desktop/Web Application as a Worked Example"). The maturity ledger lists "a Desktop/Web Application" under Exists today. No standalone Web application is live as of the original audit date.
openteams.com · 2026-08-18
Intelligence Hub deployments
Exists today — offered as an enterprise deployment
Notes & source
openteams.com offers the Intelligence Hub as its enterprise deployment tier. Revision 9's maturity ledger goes further and claims "owned Intelligence Hubs in production" under Exists today — an unquantified claim from an interested party. No deployment counts or customer numbers are published anywhere; this site must never state or imply one, and must not restate "in production" as if it were independently verified.
openteams.com · 2026-08-18
Organizational Memory
Exists today — a capability continuum, not a single product
Notes & source
Explicitly a continuum from versioned files to dedicated memory systems. The sufficiency claim revised in Revision 8 survives verbatim in Revision 9: "Many organizations will start here, but most will not find this sufficient." Revision 9 adds no maturity-ledger row for Organizational Memory as such; the nearest ledger claims are "Frames as governed artifacts with storage, identity, and connectors" (Exists today) and the accountability-plane runtime (Thesis). The continuum itself remains the safest component to describe publicly.
whitepaper §3.4; field-guide §7 · 2026-08-18
In active development
Nebari core
In active development — early access, per nebari.dev
Notes & source
The project's own label is "(early access)" as of the 2026-07-25 audit. Security, authentication, role-based access, and observability are design goals of the new core. Revision 9 does not use the name "Nebari core"; §3.1 calls the foundation layer nebari-infrastructure-core, and the paper's maturity ledger lists "Nebari and NIC" under Exists today. The live project page governs launch-time status, so the early-access label stands; note the disagreement rather than upgrading it.
nebari.dev · 2026-08-18
Nebi
In active development — early release
Notes & source
nebi.nebari.dev says "early-release" (v0.13, Apache-2.0) as of the 2026-07-25 audit. Live scope today is environment management for teams (Pixi workspaces, sync/share, diff, OCI publishing). Revision 9 raises its own claim — §3.2 says Nebi "is built and working today," built on the pixi ecosystem, which itself built on conda, and the maturity ledger lists Nebi under Exists today. The live project page still says early release and governs launch-time status, so this label holds. The paper's larger role for Nebi (packaging and versioning Frames, Cogs, Ops, and Guards across Hubs) remains roadmap by the paper's own account: Revision 9 puts the cross-Hub marketplace those artifacts travel through under Thesis.
nebi.nebari.dev · 2026-08-18
First Frames, Cogs, Ops, and Guards
In active development — the paper's own maturity ledger puts them here
Notes & source
Revision 9 self-assesses these for the first time. Its ledger puts "Cogs and Ops as first-class, installable objects with declared tools and permissions; the Op manifest with a declared validation strategy; multi-organization Hubs; the first Guard libraries" under In active development, and "Frames as governed artifacts with storage, identity, and connectors" under Exists today. Phase 1 still lists "first Ops, Cogs, Frames, and Guards built" and "Desktop/Web Application released" as milestones, so the ledger and the roadmap are in mild tension; state that rather than resolving it. Every named artifact in the guide and on this site (including the paper's vendor-fraud-review running example) is illustrative. No shipped, named Frame, Cog, Op, or Guard exists publicly as of the audit date.
whitepaper "Where This Sits in the AI Stack" maturity ledger, §10 Phase 1; field-guide/02-field-guide.md FAQ 15 · 2026-08-18
Proposed standard
Frame protocol
Proposed standard — publication is a near-term roadmap milestone
Notes & source
No published specification exists at any public URL as of the audit date. Revision 9 still lists "Frame protocol published" as a Phase 1 (now – 6 months) milestone, so publication remains ahead, not behind. The guide itself says "once published, a near-term milestone on the paper's roadmap." The guide's own success indicator: a genuinely open specification with implementations not controlled by any single company — the wording Revision 9's own de-branding invites.
field-guide/02-field-guide.md §3, §12; whitepaper §10 Phase 1 · 2026-08-18
Forecast or thesis
Public marketplace
Forecast or thesis — the paper labels it a thesis
Notes & source
Not live anywhere, and Revision 9 now says so itself: its maturity ledger puts "a marketplace exchanging Frames, Cogs, Ops, and Guards across many independent Hubs" under Thesis. Revision 9 also de-brands it — it is "the marketplace," no longer "the OpenTeams marketplace" — so this site should not call it an OpenTeams property. The guide places the public marketplace and vertical ecosystems in later phases; the nearest live analogue is nebari.dev's "Community hub (coming soon)".
whitepaper maturity ledger ("Where This Sits in the AI Stack"); field-guide/02-field-guide.md FAQ 15; nebari.dev ("Community hub — coming soon") · 2026-08-18
The distributed AI economy
Forecast or thesis — the paper labels it a thesis
Notes & source
Revision 9's maturity ledger names "the distributed AI economy itself" a thesis, alongside the accountability-plane runtime at scale. The guide's own framing keeps the same three tenses distinct — "the infrastructure layer is real and inspectable now, the execution standards are hardening, and the economy is a forecast" — but it is now the paper's self-assessment too, and the guide should re-attribute rather than claim the move. Section 12 lists falsifiable indicators: the guide's, not the company's.
whitepaper maturity ledger ("Where This Sits in the AI Stack"); field-guide/02-field-guide.md §12, FAQ 15 · 2026-08-18
Go deeper
The Distributed AI Economy: Intelligence Hubs, Frames, Cogs, Ops, and the Accountability Plane
The primary source this guide interprets: the complete architecture (Frames, Cogs, Ops, the Accountability Plane, the Intelligence Hub, and the economy thesis) in its own words.
Sources & methodology · Talk to OpenTeams · Learn more about OpenTeams