YC-13+5

A Framework for Identifying and Assessing AI-Native Companies

Research manuscript · 5 September 2026

David Quan · Michael Quan, Tutorwise Technologies

Independent research framework. “YC-13” is the authors’ research label for a synthesis of public evidence from Y Combinator-backed companies; it is not a Y Combinator taxonomy, endorsement, certification or publication.

[!NOTE] INDEPENDENCE NOTICE YC-13+5 is an independent framework. It is not authored, endorsed, certified or published by Y Combinator. “YC-13” refers to thirteen characteristics synthesized from recurring organisational properties observed in YC-backed AI-native and agent-native companies. The “+5” are proposed extensions supported by independent research and operating-system analysis.

---

Contents

Framework at a Glance

Evidence Standard

4.1 Hold-Out Application Sample

Research Propositions

Abstract

1. Research Problem

2. Terminology

3. Research Design and Selection Method

4. YC Portfolio Observation Set

5. YC-13 Derived Criteria

6. What YC-13 Captures — and What It Does Not

7. YC-13+5: Proposed Operational-Maturity Extensions

7.2 YC-13+5 Comparative Evidence Matrix

8. Category Versus Maturity: A Deliberate Open Question

9. Counterexamples and Boundary Cases

10. Assessment Method

11. Author-Affiliated Implementation Case: Tutorwise Technologies

12. Falsification Tests and Threats to Validity

13. Research Agenda

14. Conclusion

References

Appendices

Framework at a Glance

YC-13+5 separates two questions: whether an AI workforce has become part of a company’s organisational structure, and how maturely that workforce is measured, governed, audited and recovered. The framework is intentionally multidimensional: it is not a certification score and does not yet claim that all eighteen criteria are individually necessary or jointly sufficient.

ID	Canonical criterion	Operational question
YC13-01	Autonomous Execution	Can AI agents carry substantive work without step-by-step human instruction?
YC13-02	Persistent Organisational Identity/State	Does organisational state persist beyond one session or worker?
YC13-03	Agent-to-Agent Coordination	Can AI agents communicate and coordinate work directly?
YC13-04	Dynamic Organisational Structure / Delegation	Can work be delegated or routed dynamically across organisational actors?
YC13-05	Institutional Memory	Can organisational knowledge compound and be reused?
YC13-06	Explicit Authority / Roles	Are roles, responsibilities and authority boundaries explicit?
YC13-07	Governance / Human-Agent Boundaries	What is the AI agent permitted to do, and where must humans intervene?
YC13-08	Event-Driven Activation	Can work begin from events, schedules or state changes rather than prompts alone?
YC13-09	Shared Capabilities / Tools	Can the AI workforce access governed organisational capabilities?
YC13-10	Model/Vendor-Decoupled AI Roles	Can organisational roles survive a change of model/provider interface?
YC13-11	Elastic AI Workforce	Can AI work capacity expand or contract without proportional human headcount?
YC13-12	Company State Independent of Individual Workers	Does critical company state belong to the organisation rather than one actor?
YC13-13	Real-World Execution	Can AI agents execute consequential work in real systems?
YC13+5-14	Outcome Feedback Loops	Are decisions/actions linked to later outcomes and measurement?
YC13+5-15	Outcome-Calibrated Autonomy	Can measured performance and risk change the authority granted to AI agents?
YC13+5-16	Executable Governance	Do governance rules technically constrain runtime execution?
YC13+5-17	Organisational Observability, Provenance & Auditability	Can the organisation reconstruct who/what acted, under whose authority, and what happened?
YC13+5-18	Autonomous Resilience & Recovery	Can failed or missed execution be detected, preserve state, degrade safely, retry or recover, and escalate when necessary?

Figure 1. YC-13+5 two-layer architecture and organisational control loop

Abstract

“AI-native company” is increasingly used for businesses in which AI agents perform substantive organisational work, but the term lacks a widely accepted operational test. This paper presents YC-13+5, a multidimensional framework for identifying and assessing AI-native companies. The YC-13 baseline is independently synthesized from recurring organisational characteristics in a deliberately selected sample of YC-backed companies that explicitly describe themselves as AI-native, agent-native, or place AI agents in core operational roles. The framework is not a Y Combinator specification. Five additional criteria address a different problem: how an operating AI workforce is measured, governed, audited and recovered over time. These are Outcome Feedback Loops, Outcome-Calibrated Autonomy, Executable Governance, Organisational Observability, Provenance & Auditability, and Autonomous Resilience & Recovery. The paper distinguishes structural AI-nativeness from operational maturity, separates implementation strength from evidence quality, states falsification tests, and applies the frozen framework to Tutorwise Technologies as an author-affiliated implementation case. The purpose is not certification. It is to create a testable vocabulary that can be challenged by counterexamples, independent assessors and comparative evidence.

1. Research Problem

AI adoption is not the same thing as organisational redesign. A company may give every employee access to an AI assistant and remain structurally conventional: humans still hold the roles, coordinate the work, preserve the institutional context and decide when software acts. At the other extreme, a company may build an agentic product without operating itself as an AI-native company. The classification problem is therefore organisational rather than technological.

Core question: What properties distinguish a company whose operating model materially depends on an AI workforce from a company that merely uses AI?

YC-backed companies provide a useful contemporary observation set because a growing number explicitly describe themselves as AI-native or agent-native and publicly explain how AI agents perform work. Pentagon describes persistent AI agents that communicate, delegate, share context and coordinate through real-time events.[1] Wato describes shared organisational memory, reusable workflows, permissioned tools, automations and audit logs.[2] item describes AI agents sharing a company source of truth and autonomously running recurring business processes.[4] Prism describes an AI-native recruiting agency in which software and AI agents execute most of the recruiting workflow at a scale a conventional agency cannot physically match.[5]

2. Terminology

AI agent	A software actor that can pursue goals through reasoning, planning, tool use and action with a degree of autonomy.
AI workforce	A set of AI agents operating in defined roles, teams or workflows as part of an organisation’s execution capacity.
AI-native company	A company whose operating model materially depends on an AI workforce together with persistent organisational context, delegated authority, shared capabilities, governance and real-world execution. Material dependence means removing the AI workforce would require substantial reassignment of ongoing work, materially reduce operating capacity, or require redesign of the operating process—not merely remove an optional productivity tool.
AI-enabled company	A company that uses AI tools or features while its underlying operating model remains predominantly human-coordinated.

3. Research Design and Selection Method

This paper uses a structured qualitative synthesis rather than a statistical claim about the entire YC portfolio. Public YC company profiles and launch descriptions were reviewed as of 5 September 2026. Candidate companies were identified through YC company-directory searches and public YC launch material using terms including “AI-native”, “agent-native”, “AI agents”, “agents”, and descriptions of agent-run operational work. The search was purposive rather than exhaustive: this version does not claim a complete census of the YC portfolio and therefore does not report a denominator that cannot be reproduced reliably. A candidate entered the derivation sample when its public YC material satisfied at least one of four inclusion rules: (1) it explicitly described the company or operating model as AI-native; (2) it explicitly described the work environment as agent-native; (3) AI agents were described as performing substantive end-to-end organisational work rather than merely assisting a user; or (4) the product supplied organisational infrastructure specifically for persistent, coordinated AI-agent work.

Companies were not selected because they resembled Tutorwise Technologies. Tutorwise Technologies was excluded from framework derivation and introduced only after YC-13 and the proposed +5 had been defined. Public company language was coded conservatively: an observed property was recorded only when the source directly stated or closely demonstrated it. Absence of a coded property means “not established by the reviewed public material”, not “the company lacks this capability”.

Three epistemic labels are used. Observed means the property is directly described in portfolio evidence. Synthesised means multiple observed behaviours are abstracted into a broader organisational construct. Inferred means the criterion is supported by the architecture and wider agent ecosystem but is less frequently explicit in company descriptions. This distinction matters because a framework should not make every criterion look equally evidenced when it is not. A second coding dimension distinguishes operating-company evidence from AI-native infrastructure evidence. Operating-company evidence describes a company whose own delivery model materially places AI agents in organisational work. Infrastructure evidence describes products that make persistent, coordinated or governed AI-agent work possible for other organisations. The latter can establish an architectural property without proving that the infrastructure provider itself operates as an AI-native company.

3.1 Evidence Standard

To prevent persuasive narrative from being mistaken for proof, YC-13+5 uses two separate dimensions: provenance of the criterion and strength of the evidence used to assess an implementation. The latter should be read as an evidence ladder, not as a maturity score.

Code	Evidence level	Meaning
E0	Assertion only	Claim is made but no inspectable supporting evidence is supplied.
E1	Publicly documented	A public source describes the capability; implementation is not independently inspected.
E2	Implementation inspected	Code, configuration, schema or equivalent implementation evidence is inspected.
E3	Operationally corroborated	Runtime records or live operating evidence corroborate that the mechanism operates in practice.
E4	Independently verified	An independent assessor verifies the implementation and/or operation under a defined protocol.

Tutorwise Technologies does not receive E4 merely because its code has been inspected in this study; the case remains author-affiliated and requires independent replication for that designation.

4. YC Portfolio Observation Set

Company	Public positioning	Observed organisational evidence	Ref.	Evidence class
Pentagon	Agent-native coordination layer	Persistent agents; agent-to-agent messaging; delegation; event-driven coordination; shared skills/tools; human-in-the-loop	[1]	Infra.
Wato	Shared AI workspace	Institutional memory; company-held context; permissioned tools; automations; audit logs; multiple AI environments	[2]	Infra.
Korso	AI-native agency / Shepherd	Cross-session and cross-team agent communication; collaboration with zero human coordination input	[3]	Operating
item	AI-native CRM	Company source of truth; autonomous recurring business processes; agents trained from process documents; real actions	[4]	Operating
Prism	AI-native recruiting agency	End-to-end recruiting work; dozens of searches in parallel; human time reserved for interviews/offers	[5]	Operating
Perfectly	AI-native recruiting OS	Automates sourcing, outreach, screening and qualification; recruiting agent operates candidate workflow	[6]	Operating
Minerva	AI-native accounting firm	Accountants collaborate with intelligent agents; agents ingest, follow up, reconcile and continuously forecast	[7]	Operating
Foaster	AI-native consulting firm	AI agents perform scalable consulting work; experts review outputs; corrections feed back into the system	[8]	Operating
flowscope	AI-native transformation consulting	Agents map, optimise and automate business processes end-to-end and ship working software	[9]	Operating
Peer	AI-native freight brokerage	Agents quote, vet carriers, handle paperwork, track delivery and resolve problems	[10]	Operating
Panta	AI-native insurance brokerage	AI operators use real systems, forms, portals and calls to perform broker work across many clients	[11]	Operating
Wealor	AI-native wealth-management platform	Knowledge layer plus specialised agents executing operational work across legacy systems	[12]	Operating
Marker	Agent-first consultancy	Agents deployed into production; each agent deepens business understanding for subsequent agents	[13]	Operating
Dialogus	AI-native contact centre	Production agents integrate with enterprise systems, complete workflows and escalate edge cases to humans	[14]	Operating
Fixture	AI-native CRM	Unified context store; proactive AI agents; access through multiple AI tools	[15]	Infra.
Corvera	Context layer for AI-native brands	Any AI tool via MCP; workflow automation; access control and audit logging	[16]	Infra.

4.1 Hold-Out Application Sample

As a small non-derivation check, v1.0 applies the frozen framework to four additional YC-backed companies that were not part of the original sixteen-company derivation set. The purpose is not statistical validation; it is to test whether the framework can describe new cases without changing the criteria.

Hold-out company	Public positioning	Framework properties visible in public evidence	Result
Locke	AI-native government affairs firm	AI agents perform substantive government-affairs work alongside policy staff; real-world operating model evidence.	Fits Layer I directionally; public evidence is insufficient to score all YC-13 criteria.
Hedge	AI-native specialty insurance company	Workflow rebuilt around AI agents paired with underwriting; operating-company evidence rather than an AI feature alone.	Fits AI-native operating-model boundary; detailed persistence/governance evidence not public.
Trope	AI-native platform for ERP services	Agents perform project-manager/consultant/developer work, act on ERP implementation state and keep artefacts/stakeholders synchronised.	Supports autonomous execution and real-world workflow redesign; not enough public evidence for full YC-13 scoring.
Terminal Use	AI-native transformation / background-agent infrastructure	Long-running agents, persistent messages/filesystems, schedules, multi-agent configurations, framework-agnostic hosting and granular permissions.	Strong infrastructure corroboration for persistence, events, decoupling, coordination and governance.

Hold-out finding: no criterion was added, removed or reworded to accommodate these four cases. The exercise therefore provides a modest out-of-sample coherence check, not independent validation.

Interpretive weighting of source types. Infrastructure evidence was used to establish the recurrence and architectural feasibility of organisational properties. Operating-company evidence was given greater interpretive weight when judging whether a property characterised an AI-native operating model, because infrastructure capability alone does not establish that the provider itself operates as an AI-native company.

5. YC-13 Derived Criteria

The thirteen baseline criteria below are frozen for Version 1.0 before the Tutorwise Technologies case is assessed. They are not claimed to be necessary and sufficient conditions. Rather, they are recurring organisational properties that collectively distinguish an AI-native operating model from ordinary AI adoption.

#	Criterion	Operational test	Provenance
1	Autonomous execution	AI agents can carry work forward without a human coordinating every step.	Observed
2	Persistent organisational identity and state	Roles, responsibilities, state and context persist beyond a single prompt, session or model invocation.	Synthesised
3	Agent-to-agent coordination	AI agents can communicate, share context and coordinate work directly.	Observed
4	Dynamic organisational structure and delegation	Work can be delegated and routed among AI agents according to objective, task state or required capability.	Observed
5	Institutional memory	Useful organisational knowledge persists and can be reused by future AI agents and human colleagues.	Observed
6	Explicit authority and roles	AI agents operate within defined roles, responsibilities and authority boundaries.	Synthesised
7	Governance and human-agent boundaries	Human and AI decision rights are distinguished, with escalation or approval boundaries for consequential actions.	Observed
8	Event-driven activation	AI agents and workflows can be activated by organisational events or cadence, not only by a human prompt.	Observed
9	Shared capabilities and tools	AI agents can inherit or access governed tools, skills and integrations at organisational scope.	Observed
10	Model/vendor-decoupled AI roles	The organisational role, authority, state and capabilities are not inseparably bound to one model provider or agent interface.	Inferred / increasingly observed
11	Elastic AI workforce	The organisation can vary AI execution capacity according to workload rather than treating every AI agent as a fixed seat.	Synthesised
12	Company state independent of individual workers	Critical organisational state is held by the company rather than disappearing with an individual human, AI-agent session or model.	Synthesised
13	Real-world execution	AI agents can take consequential actions in business systems and workflows, not merely generate recommendations or text.	Observed

5.1 Evidence audit: what survived scrutiny

The expanded portfolio review strengthens the baseline but also changes how confidently individual factors should be described. Criteria 1, 3, 5, 7, 8, 9 and 13 have unusually direct public support. Pentagon alone provides explicit evidence for coordination, delegation, persistence, shared capabilities, event-driven communication and human-in-the-loop operation.[1] Wato independently supports memory, company-held context, permissions, automations and auditability.[2] item, Prism, Peer, Panta and Dialogus provide strong evidence that real-world execution is central to the emerging category.[4][5][10][11][14]

Criterion 10 was the principal concern in v0.2. The wider audit does not justify removing it, but it does justify a more careful provenance label. Wato explicitly works across Codex, Claude Desktop, Cursor, Claude Code and other MCP-compatible environments; Fixture can be used through different AI tools; and Corvera is designed to make organisational context legible to any AI tool through MCP.[2][15][16] These are meaningful signals that organisational context and capability are being separated from a single model interface. They do not prove that model/vendor independence is already universal among AI-native companies. Criterion 10 therefore remains in YC-13 as an inferred and increasingly observed organisational property, explicitly exposed to falsification.

Criteria 11 and 12 are similarly stronger as organisational syntheses than as phrases used by founders. Prism’s parallel search capacity and Peer’s claim that agents can handle far more loads with far fewer people support elastic execution.[5][10] Wato and item directly support the idea that company context and knowledge should remain with the organisation rather than individual chats or users.[2][4] The conceptual abstraction—elastic AI workforce and company-held state—is ours.

6. What YC-13 Captures — and What It Does Not

YC-13 primarily describes Layer I — AI-Native Organisational Structure: whether the company can field an AI workforce that acts, persists, coordinates, delegates, remembers, uses shared capabilities, operates under authority boundaries and executes real work. The +5 are proposed as Layer II — AI-Native Operational Maturity: how that workforce is measured, governed, audited and recovered once entrusted with persistent authority and real-world work. This two-layer interpretation is a research hypothesis, not a certification rule.

Once AI agents are entrusted with ongoing work, five managerial questions become unavoidable. Did their decisions produce the intended outcome? Should demonstrated performance change how much autonomy they receive? Are governance boundaries technically enforceable? Can the company reconstruct who or what acted and under whose authority? Can the organisation detect and recover from failed execution? Contemporary research increasingly treats autonomy as graded rather than binary, governance as architectural, observability as a reliability requirement, and evaluation loops as part of production agent operation.[17–21]

7. YC-13+5: Proposed Operational-Maturity Extensions

#	Criterion	Definition	Distinctive test
14	Outcome Feedback Loops	The organisation links AI-agent decisions and actions to later observable outcomes.	Epistemic loop: Decision → Action → Outcome → Measurement. Memory records what happened; this measures whether it worked.
15	Outcome-Calibrated Autonomy	The autonomy granted to an AI agent, role or process can expand, contract or remain constrained according to measured performance and risk.	Authority loop: Measurement → Confidence/Risk → Authority adjustment. Governance defines boundaries; calibration makes autonomy evidence-responsive.
16	Executable Governance	Authority, approval, risk and escalation rules materially constrain runtime execution rather than existing only in documents, prompts or training.	Policy-as-code is one implementation, but runtime gates, permissions, graph interrupts, transaction limits and capability restrictions also qualify.
17	Organisational Observability, Provenance & Auditability	The company can reconstruct material AI-workforce activity, including actor/seat, authority, delegation, tools, action, status, cost and outcome where relevant.	Persistence means state exists. Observability and provenance make consequential activity inspectable and attributable.
18	Autonomous Resilience & Recovery	The AI workforce can detect failed or missed execution, preserve state, degrade safely, retry or recover, and escalate when necessary.	Autonomous execution describes acting when things work; resilience describes organisational behaviour when they do not.

7.1 Independent support

OpenAI Frontier provides a useful industry comparator because it explicitly combines durable institutional memory, production agent execution, built-in evaluation and optimisation loops, explicit permissions, auditable actions and observability.[17] This does not validate YC-13+5, but it independently demonstrates that outcome evaluation, governance and observability are converging as production requirements for enterprise AI agents.

Kasirzadeh and Gabriel characterise AI agents across graded dimensions including autonomy and argue that different agentic profiles create different governance challenges.[18] Dux and colleagues, using an enterprise case, argue that governance is implemented through concrete architectural and working arrangements determining what an agent may do, which tools and data it can use, how memory is handled and how performance improvements are introduced.[19] Together these support the logic behind Outcome-Calibrated Autonomy and Executable Governance.

SAGA provides a stronger security antecedent for executable governance and provenance by implementing user-defined access-control policies and cryptographically derived controls over inter-agent interaction.[20] AgentTelemetry and recent federated-observability research treat telemetry, fault detection, heterogeneous agent events and governance-oriented analysis as first-class concerns in deployed agentic systems.[21][22] These sources support #17 and, indirectly, the need for #18: systems that cannot detect and diagnose failure cannot reliably recover from it.

Direct support for #18 is provided by AgentRewind, a 2026 runtime-recovery framework for long-horizon LLM agents. It records aligned checkpoints of agent context and controlled environment state, allows rollback to an earlier state, and resumes execution with information from prior attempts; experiments report improved task success and partial-progress outcomes across models and harnesses.[23] This is closer to the resilience claim than observability alone because it treats post-failure recovery as an explicit runtime capability.

7.2 YC-13+5 Comparative Evidence Matrix

This matrix makes the derivation/assessment sequence explicit. For criteria 1–13, the middle column identifies YC portfolio evidence supporting the organisational property. For criteria 14–18, which are proposed extensions rather than YC-derived criteria, the same column shows YC portfolio corroboration where it genuinely exists and independent research where it does not. A qualified statement is preferable to implying evidence that has not been established. Tutorwise Technologies is assessed only in the third column.

Criterion	YC portfolio company evidence / external corroboration	Tutorwise Technologies
1. Autonomous execution	Pentagon; item; Prism; Peer; Panta; Dialogus — agents carry substantive work without step-by-step human coordination.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — SpecialistAgentRunner, TeamRuntime and event-driven workflow execution.
2. Persistent organisational identity/state	Pentagon; Wato; item; PathPilot — persistent sessions, company memory, shared context and continuing operational state.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — DB-backed agent/team state plus PostgresSaver checkpoints.
3. Agent-to-agent coordination	Pentagon; Korso; PathPilot — direct messaging, cross-agent collaboration and coordinated AI workforces.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — organisational bus with persisted requests, tasks, decisions, reports and handoffs.
4. Dynamic organisational structure/delegation	Pentagon — agents message one another, delegate tasks and operate within defined organisational structure.	STRONG · E2 IMPLEMENTATION INSPECTED — supervisor/pipeline/swarm routing within loaded team topology; not unrestricted organisation-wide composition.
5. Institutional memory	Pentagon; Wato; item; PathPilot — self-evolving memory, shared company context and reusable operational knowledge.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — episodic/factual memory, semantic retrieval and fact invalidation.
6. Explicit authority and roles	Pentagon; item; Agentic Fabriq; PathPilot — defined agents/roles, scoped permissions and workflow-specific responsibilities.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — role definitions, ITSM risk classes and human-reserved high-risk actions.
7. Governance / human-agent boundaries	Pentagon; Wato; Agentic Fabriq — human-in-the-loop operation, permissioning, approval flows and least-privilege controls.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — LangGraph HITL interrupt/resume plus RFC and ITSM approval gates.
8. Event-driven activation	Pentagon; Wato; item — realtime events, connector triggers, cadence and recurring autonomous business processes.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — webhook, database and cron activation with missed-event recovery.
9. Shared capabilities/tools	Pentagon; Wato; item; Agentic Fabriq — shared skills, approved skills and governed integrations and governed capability access.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — central tool registry,
10. Model/vendor-decoupled AI roles	Wato; Fixture; Corvera — organisational memory/tools/context exposed across multiple AI environments rather than one model interface.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — configurable providers/models; company-held state and bus are not bound to one provider. Current operations use multiple vendor seats.
11. Elastic AI workforce	Prism; Peer; Panta — parallel agent work and operational capacity that scales without proportional human headcount.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — workload-driven worker fan-out with configured lane capacity and self-throttles.
12. Company state independent of individual workers	Wato; item; PathPilot — company context persists across agents, users and workflow stages rather than living in one session.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — DB-held roles, team config, memory, workflow state and checkpoints survive process replacement.
13. Real-world execution	item; Prism; Peer; Panta; Dialogus — agents run business processes, transact with systems and complete operational work.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — write tools and ITSM integrations create and transition real external work.
14. Outcome Feedback Loops	PathPilot explicitly describes an Audit & Quality Agent and continuous feedback loop; Foaster describes expert corrections feeding the system. Independent support: OpenAI Frontier.	STRONG · E2 IMPLEMENTATION INSPECTED — decision_outcomes plus workflow-completion hooks create records supporting subsequent outcome assessment; recurring linkage to later observed business outcomes remains to be operationally corroborated.
15. Outcome-Calibrated Autonomy	No direct YC portfolio evidence yet establishes closed-loop authority adjustment. Independent support: Kasirzadeh & Gabriel; Dux et al.	STRONG · E2 IMPLEMENTATION INSPECTED — process_autonomy_config stores tiers, accuracy, thresholds and proposals; closed-loop production promotion/demotion remains to be behaviourally verified.
16. Executable Governance	Agentic Fabriq; Wato; Corvera — scoped permissions, approval flows, policy control and governed tool access. Independent support: SAGA.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — Human-reserved ITSM changes, RFC approval gate and graph-level HITL interrupt materially constrain execution.
17. Organisational Observability, Provenance & Auditability	Agentic Fabriq; Wato; PathPilot — central action logs, audit trails, tool-call tracing and cross-system quality review. Independent support: AgentTelemetry.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — agent_telemetry plus bus provenance and usage/cost/status records.
18. Autonomous Resilience & Recovery	No direct YC portfolio evidence is yet treated as sufficient. Independent support: AgentRewind checkpoint, rollback and resumed long-horizon execution.	VERY STRONG · E2 IMPLEMENTATION INSPECTED — Failed-webhook retry, missed-event recovery, checkpoint persistence and held/escalated failure states.

Interpretation note: this is an evidence matrix, not an endorsement matrix or a composite score. YC portfolio mentions establish public evidence for a property, not that every named company satisfies the full criterion. Criteria 14–18 remain proposed YC-13+5 extensions.

8. Category Versus Maturity: A Deliberate Open Question

The audit surfaced a distinction that v0.2 treated too weakly. YC-13 and +5 should not automatically be treated as eighteen equal membership tests. The working hypothesis is that Layer I — AI-Native Organisational Structure (YC-13) primarily describes the structure of an AI-native company, while Layer II — AI-Native Operational Maturity (+5) primarily describes the maturity with which that company manages an AI workforce.

H1 — YC-13 primarily identifies AI-native organisational structure. H2 — +5 primarily measures the maturity with which that AI-native organisation is managed.

This distinction is intentionally not resolved by assertion. Comparative testing may show that some YC-13 criteria are maturity properties, or that some +5 factors become necessary conditions once the AI workforce crosses a threshold of autonomy or consequence. A company should therefore not be declared “not AI-native” merely because it lacks sophisticated recovery or observability. It may instead be an immature AI-native company. Conversely, excellent observability around a conventional chatbot does not make the company AI-native.

8.1 Criterion-Distinctness Note

Three criterion clusters are intentionally adjacent but non-equivalent. YC13-02 tests persistence through execution/session boundaries: does organisational state survive over time? YC13-05 tests knowledge accumulation and reuse: can experience and context become reusable organisational knowledge? YC13-12 tests organisational ownership and worker substitutability: does critical company state remain available when a particular worker, model or session disappears? Likewise, YC13-06 tests whether roles and authority are represented; YC13-07 tests whether human and AI decision rights and boundaries are defined; YC13+5-16 tests whether those governance rules can technically stop or constrain execution.

9. Counterexamples and Boundary Cases

Case	Description	Framework interpretation
AI everywhere, conventional company	Every employee uses ChatGPT, Claude or Copilot, but humans still own roles, coordination, institutional state and execution.	AI-enabled, not established as AI-native.
Agentic product	A company sells an autonomous customer-support agent but internally operates like a conventional SaaS company.	Agentic product ≠ AI-native company.
Single autonomous workflow	One back-office process runs autonomously but has no persistent organisational roles, cross-agent coordination or company-held state.	Strong automation; insufficient evidence of an AI-native operating model.
Persistent AI workforce	AI agents occupy defined roles, coordinate, share institutional context, operate under authority boundaries and execute work across business systems.	Strong AI-native candidate under YC-13.
Mature AI-native organisation	The previous case plus measured outcome loops, calibrated autonomy, executable governance, provenance and recovery.	Strong candidate for high YC-13+5 operational maturity.

9.1 Boundary Archetypes

Archetype	AI-native assessment	Reason
Employees use general AI assistants	No	AI assists human workers; the organisational operating model remains human-held.
SaaS product includes one AI chatbot	No	An AI feature is not an AI workforce or organisational architecture.
One autonomous functional agent, no persistent organisational architecture	Borderline / partial	Agentic execution exists, but company-level persistence, coordination and authority may not.
Persistent AI agents hold roles, coordinate and execute business processes	Strong candidate	The architecture of work has changed, not merely the software interface.
Same organisation plus feedback, calibrated autonomy, executable governance, observability and recovery	Mature AI-native organisation	Layer II capabilities manage the AI workforce as an operating organisation.

10. Assessment Method

YC-13+5 v1.0 is a multidimensional assessment framework, not a certification. It does not yet specify a numerical threshold such as “12/13”. A future necessary-condition study should test whether autonomous execution, persistent company-held state, AI-agent coordination, authority/governance, shared capabilities and real-world execution form an indispensable minimum core. Until that evidence exists, scores should describe evidence and maturity rather than confer status. The assessment separates implementation strength from evidence quality so that a persuasive narrative cannot substitute for an operating mechanism.

Dimension	Labels	Interpretation
Implementation strength	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	How fully the organisational property is implemented.
Evidence quality	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED	How directly the claim is supported. Current executable code is stronger than descriptive documentation for implementation claims.
Framework status	STRUCTURAL / MATURITY / OPEN	A research label, not a score: whether future evidence supports the factor as category-defining, maturity-related or unresolved.

10.1 Necessary, Sufficient and Aggregate-Score Limits

YC-13+5 does not currently claim that all eighteen criteria are individually necessary or jointly sufficient for AI-native status. It also does not establish equal criterion weights or a validated aggregate measure. Accordingly, criterion ratings must not be summed into a percentage, league table or certification score. Future work should identify necessary conditions empirically and test whether Layer II maturity predicts reliability, safe delegation, recovery, cost efficiency or organisational performance.

10.2 What YC-13+5 Is Not

YC-13+5 is not a Y Combinator framework or endorsement; not a ranking of YC companies; not a certification scheme; not a measure of whether a company is commercially successful or ethically superior; not a measure of model intelligence; not a claim that humans disappear from organisations; and not an argument that more autonomy is always better.

11. Author-Affiliated Implementation Case: Tutorwise Technologies

Tutorwise Technologies is introduced only after the framework has been frozen. It is an author-affiliated participant implementation case, not independent validation: the assessed organisation participated in developing the framework. The case therefore tests implementation and exposes weaknesses rather than defining the criteria. The evidence hierarchy is: current executable code as primary evidence; database migrations and runtime configuration as implementation evidence; live operational records as corroboration; and documentation as explanation rather than proof.

LIVE OPERATING EVIDENCE At the 5 September 2026 snapshot,
the Tutorwise Technologies AI Company Dashboard exposed repository,
ticketing, publishing, financial, governance and multi-vendor
AI-workforce measurements, including adverse or incomplete readings and
explicit distinctions between measured and modelled figures.[24] Open
the AI Company Dashboard The dashboard is supporting evidence, not an
independent audit.

---

11.1 Criterion-by-criterion implementation assessment

#	Criterion	Strength	Evidence	Primary implementation evidence
1	Autonomous execution	VERY STRONG	E2 IMPLEMENTATION INSPECTED	SpecialistAgentRunner ReAct/tool execution; TeamRuntime; webhook/cron execution.
2	Persistent organisational identity and state	VERY STRONG	E2 IMPLEMENTATION INSPECTED	DB-backed specialist agents/team topology plus PostgresSaver checkpoints.
3	Agent-to-agent coordination	VERY STRONG	E2 IMPLEMENTATION INSPECTED	writeBusMessage → agent_org_handoff; request/decision/task/report messaging and persisted handoffs.
4	Dynamic organisational structure and delegation	STRONG	E2 IMPLEMENTATION INSPECTED	TeamRuntime supports supervisor/pipeline/swarm patterns and dynamic NEXT_AGENT routing within loaded team topology; not yet unrestricted organisation-wide capability composition.
5	Institutional memory	VERY STRONG	E2 IMPLEMENTATION INSPECTED	AgentMemoryService persists episodes/facts, retrieval and fact invalidation.
6	Explicit authority and roles	VERY STRONG	E2 IMPLEMENTATION INSPECTED	ITSM risk classification, human-reserved high-risk changes and machine-checkable RFC approval paths.
7	Governance and human-agent boundaries	VERY STRONG	E2 IMPLEMENTATION INSPECTED	LangGraph interrupt/resume HITL plus approval-gated change paths.
8	Event-driven activation	VERY STRONG	E2 IMPLEMENTATION INSPECTED	Database/webhook workflow triggers, cron execution and missed-event recovery.
9	Shared capabilities and tools	VERY STRONG	E2 IMPLEMENTATION INSPECTED	Central agent-tool registry, skills and governed execution surfaces.
10	Model/vendor-decoupled AI roles	VERY STRONG	E2 IMPLEMENTATION INSPECTED	Agent role execution supports configurable providers/models; organisational bus and company-held state are not bound to one provider. Current operations use multiple vendor seats.
11	Elastic AI workforce	VERY STRONG	E2 IMPLEMENTATION INSPECTED	Content dispatch fans out workers to workload up to configured lane capacity and self-throttles.
12	Company state independent of individual workers	VERY STRONG	E2 IMPLEMENTATION INSPECTED	DB-held roles, team config, memory, workflow state and checkpoints survive process replacement.
13	Real-world execution	VERY STRONG	E2 IMPLEMENTATION INSPECTED	Write tools and ITSM integrations create/transition real external work.
14	Outcome Feedback Loops	STRONG	E2 IMPLEMENTATION INSPECTED	decision_outcomes plus workflow-completion hooks create records supporting subsequent outcome assessment; recurring linkage to later observed business outcomes remains to be operationally corroborated.
15	Outcome-Calibrated Autonomy	STRONG	E2 IMPLEMENTATION INSPECTED	process_autonomy_config stores tiers, accuracy, thresholds and proposals; closed-loop production promotion/demotion remains to be behaviourally verified.
16	Executable Governance	VERY STRONG	E2 IMPLEMENTATION INSPECTED	Human-reserved ITSM changes, RFC approval gate and graph-level HITL interrupt materially constrain execution.
17	Organisational Observability, Provenance & Auditability	VERY STRONG	E2 IMPLEMENTATION INSPECTED	agent_telemetry plus bus provenance and usage/cost/status records.
18	Autonomous Resilience & Recovery	VERY STRONG	E2 IMPLEMENTATION INSPECTED	Failed-webhook retry, missed-event recovery, checkpoint persistence and held/escalated failure states.

11.2 Negative evidence retained

The case is intentionally not presented as a perfect score. Criterion #4 is assessed STRONG rather than VERY STRONG because current dynamic routing occurs within loaded team topology; the stronger form—open-ended organisation-wide capability discovery, temporary composition and dissolution—is not yet established by the reviewed implementation.

A second limitation concerns provenance. The organisational bus persists sender attribution and runtime context, but the reviewed HTTP path explicitly notes that the claimed sender seat is caller-supplied under shared authentication rather than cryptographically proving ownership of that seat. This does not negate agent-to-agent coordination, but it means sender attribution should not be described as cryptographically authenticated provenance. The limitation is exactly the kind of distinction criterion #17 is intended to expose.

The outcome-calibrated autonomy implementation also warrants precise wording. Current code clearly represents autonomy tiers, accuracy metrics, thresholds and proposals and makes these values available to runtime tooling.[25] This establishes a code-verified calibration mechanism, but not yet a fully demonstrated closed behavioural loop in which production authority reliably rises or falls because of measured outcomes. For that reason #15 is rated STRONG rather than VERY STRONG in this version. It should be upgraded only when operational evidence establishes outcome-driven tier transitions.

Criterion #17 interpretation. Organisational Observability, Provenance & Auditability requires reconstructable attribution, authority provenance and action history. Version 1.0 does not require cryptographic proof of actor identity unless that stronger claim is explicitly made; accordingly, the Tutorwise bus limitation around caller-supplied sender attribution remains a disclosed negative finding.

12. Falsification Tests and Threats to Validity

Test	What would count against the framework?	Required response
Portfolio-selection bias	A broader or independently selected AI-native sample produces materially different recurring properties.	Rebuild YC-13 from the new sample rather than defend the old list.
Necessity	Credible AI-native companies repeatedly fail the same YC-13 criterion while remaining substantively AI-native.	Move that factor to maturity or remove it.
Distinctness	Independent assessors cannot reliably distinguish two criteria in real systems.	Merge or redefine them.
Criterion 10	Model/vendor independence does not recur as AI-native organisations mature and proves unrelated to organisational continuity.	Remove or reclassify #10.
YC-13 vs +5	The +5 recur as basic category requirements rather than maturity properties, or YC-13 factors behave like maturity measures.	Abandon the two-layer hypothesis.
Evidence reliability	Independent reviewers reach materially different ratings from the same code and operating evidence.	Tighten rubrics and introduce explicit evidence thresholds.
Outcome relevance	Higher framework maturity shows no reliability, safe autonomy, cost efficiency or organisational performance.	Do not claim predictive relationship to value.
Author-affiliated implementation dependence	The +5 are compelling only in Tutorwise Technologies and fail to recur elsewhere.	Reject or narrow the affected additions.

Additional threats include public-description bias: YC profiles are founder-authored and may omit capabilities or use “AI-native” as positioning rather than a technical construct. The sample also overrepresents young technology companies and service businesses. The framework therefore cannot yet claim population validity. The Tutorwise Technologies case is participant evidence from a company involved in developing the framework, creating an obvious risk of confirmation bias. Freezing criteria before case assessment, retaining negative findings and requiring current code evidence reduce but do not eliminate that risk.

13. Research Agenda

The next research stage should convert YC-13+5 from a reasoned framework into a reproducible assessment instrument. That requires a larger sample including non-YC companies, explicit coding rules, blinded independent assessors, inter-rater reliability testing and pre-registered hypotheses about which criteria are structural versus maturity-related. A useful design would compare three populations: conventional AI-enabled companies, companies selling agentic products but operating conventionally, and companies whose own operating models materially depend on an AI workforce.

Three empirical questions are especially important. First, which minimum subset of YC-13 is necessary to distinguish AI-native operating models? Second, does +5 maturity predict reliability, safe delegation, recovery, cost efficiency or organisational performance? Third, does outcome-calibrated autonomy produce a measurable advantage over static autonomy policies without increasing unacceptable risk? Until these questions are tested, YC-13+5 should remain a working framework rather than a standard.

13.1 Research Propositions

P1. AI-native companies exhibit persistent organisational state that is not reducible to an individual human or AI worker session.

P2. AI-native organisational structure can be distinguished from ordinary AI-enabled tool use by observable changes in roles, coordination, authority, state and execution.

P3. YC-13 primarily measures AI-native organisational structure rather than operational maturity.

P4. The +5 primarily measures operational maturity rather than category membership.

P5. Outcome-calibrated autonomy is associated with safer and more reliable expansion of AI-agent authority than static autonomy policies, conditional on effective governance and observability.

14. Conclusion

The central claim of YC-13+5 is deliberately narrower than the phrase “AI will transform companies”. An AI-native company should be identifiable in the architecture of work: AI agents occupy meaningful roles, carry work autonomously, coordinate, preserve institutional context, operate under authority boundaries, access shared capabilities and execute consequential workflows. Those properties describe an organisation, not merely a software feature. In this paper, material dependence means that removing the AI workforce would require substantial reassignment of ongoing organisational work, materially reduce operating capacity, or require redesign of the company’s operating process; removal of an optional productivity tool is insufficient.

The audit also suggests that structural AI-nativeness and operational maturity should be separated. YC-13 asks whether an AI workforce has become part of the company’s operating substrate. The +5 ask whether that workforce can be measured, trusted, governed, audited and recovered as an organisational system. The distinction is a hypothesis, not a conclusion.

The framework should earn credibility by surviving attempts to break it: counterexamples, independent scoring, broader samples and evidence that contradicts its authors.

For that reason, YC-13+5 v1.0 is released for public use and external criticism: specific enough to test, explicit about its provenance, and unfinished by design.

Publication, Versioning and Suggested Citation

Framework version: YC-13+5 v1.0. Paper version: 1.0. Evidence snapshot: 5 September 2026. Versioning rule: v1.x may update evidence, clarifications and examples without renumbering the canonical criterion IDs. Any change to criterion membership or numbering should trigger a major framework version.

Suggested citation:

Quan, D., & Quan, M. (2026). YC-13+5: A Framework for Identifying and Assessing AI-Native Companies (Version 1.0). Tutorwise Technologies. 5 September 2026. DOI/persistent identifier: to be assigned if deposited in a public repository.

Competing Interests and Author Affiliation

David Quan and Michael Quan are co-founders of Tutorwise Technologies, which is assessed in this paper as an author-affiliated implementation case. Tutorwise Technologies was excluded from framework derivation. The Tutorwise implementation case must not be treated as independent validation of YC-13+5.

Author Contributions

David Quan and Michael Quan jointly developed the research framing, framework interpretation and manuscript. The authors jointly reviewed the evidence synthesis and approved Version 1.0 for public release.

References

A. Primary YC portfolio evidence

[1] Pentagon — The control plane for agent-native work. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/pentagon

[2] Wato — The control point for AI agents at work. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/wato

[3] Korso — The AI-native agency. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/korso

[4] item — The AI-Native CRM that works for you. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/item

[5] Prism — AI-native recruiting agency. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/tryprism

[6] Perfectly — The AI-native Recruiting OS. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/perfectly

[7] Minerva — AI native accounting firm. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/minerva

[8] Foaster — The AI-native McKinsey. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/foaster

[9] flowscope — AI-native consulting to map and automate business processes. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/flowscope

[10] Peer — AI-native freight brokerage. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/peer

[11] Panta — AI Native Commercial Insurance Brokerage. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/panta

[12] Wealor — AI-native platform for wealth managers. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/wealor

[13] Marker — Platform and FDEs for rebuilding businesses agent-first. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/marker

[14] Dialogus — The AI-native contact center. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/dialogus

[15] Fixture — An AI-first CRM built for Startups. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/fixture

[16] Corvera — The context layer for AI-native CPG brands. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/corvera

B. Industry and independent research evidence

[17] OpenAI Frontier — Enterprise platform for AI agents. OpenAI. https://openai.com/business/frontier/

[18] Characterizing AI Agents for Alignment and Governance. Kasirzadeh & Gabriel. https://arxiv.org/abs/2504.21848

[19] Governance by Design: Architecting Agentic AI for Organizational Learning and Scalable Autonomy. Dux et al.. https://arxiv.org/abs/2605.20210

[20] SAGA: A Security Architecture for Governing AI Agentic Systems. NDSS Symposium 2026. https://www.ndss-symposium.org/ndss-paper/saga-a-security-architecture-for-governing-ai-agentic-systems/

[21] AgentTelemetry: A Fault Detection Benchmark and Toolkit for LLM Agent Observability. ACM AIware 2026. https://doi.org/10.1145/3805760.3814931

[22] A federated observability architecture pattern for reliable agentic AI software systems across the AI SDLC. Information and Software Technology. https://www.sciencedirect.com/science/article/pii/S0950584926002491

[23] AgentRewind: Recoverable Execution for Long-Horizon LLM Agents. Zhuang et al., arXiv:2608.14380 (2026). https://arxiv.org/abs/2608.14380

[26] Agentic Fabriq — The control plane for AI agents. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/agentic-fabriq

[27] PathPilot — AI Agents for Lending Operations. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/pathpilot

[28] Locke — AI agents that help companies and governments work together. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/locke

[29] Hedge — AI-Native insurance company. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/hedge

[30] Trope — AI FDE that deploys custom AI agents into ERPs. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/trope

[31] Terminal Use — AI-native transformation for operations-heavy companies. Y Combinator. Accessed 5 September 2026. https://www.ycombinator.com/companies/terminal-use

C. Tutorwise Technologies operating and implementation evidence

[24] AI-Native Company Operations Dashboard and Metrics. Tutorwise Technologies. Accessed 5 September 2026. https://www.tutorwise.io/ai-company-dashboard

[25] Tutorwise Technologies implementation evidence. Current repository snapshot reviewed at commit c657d537b10a200be824d75f7f8869a02127d483, including process_autonomy_config, TeamRuntime, writeBusMessage, SpecialistAgentRunner, workflow outcome hooks and agent_telemetry. This implementation assessment is fixed to repository commit c657d537; future framework releases should refresh the evidence snapshot against the applicable repository state.

Appendix A — Tutorwise Technologies Evidence Map

Property	Current code path	What it establishes
Autonomy calibration	apps/web/src/app/api/admin/conductor/autonomy/route.ts; tools/database/migrations/377_create_learning_loop.sql	Stores current tier, 30-day accuracy, threshold, proposals; tiers supervised / semi-autonomous / autonomous.
Outcome loop	apps/web/src/lib/conductor/workflow/runtime/PlatformWorkflowRuntime.ts	Writes decision_outcomes records after completed workflow execution to support subsequent outcome assessment; recurring linkage to later observed business outcomes remains to be operationally corroborated.
Team persistence & HITL	apps/web/src/lib/conductor/workflow/team-runtime/TeamRuntime.ts	LangGraph StateGraph, PostgresSaver, interrupt/Command and resumable human approval.
Organisational bus	apps/web/src/lib/conductor/bus/writeBusMessage.ts	Single write path for persisted agent_org_handoff messages and provenance.
Bus identity caveat	apps/web/src/app/api/conductor/bus/route.ts	HTTP route documents that from_agent/session_id are caller-supplied under shared authentication.
Provider separation	apps/web/src/lib/agent-tool-registry/SpecialistAgentRunner.ts	Agent model/provider can be configured without changing the runner.
Telemetry	tools/database/migrations/573_create_agent_telemetry_and_decisions.sql; apps/web/src/lib/agent-telemetry/index.ts	Records token usage, cost/status and operational telemetry.

Appendix B — Public Evidence Dataset

This compact dataset records the public evidence frame used in the derivation and hold-out checks. It is intended to make recoding and disagreement possible. All mutable web sources were accessed on 5 September 2026.

Company	Sample role	Evidence type	Primary public source
Fixture	Derivation	Infrastructure	Y Combinator profile
Wato	Derivation	Infrastructure	Y Combinator profile
Corvera	Derivation	Infrastructure	Y Combinator profile
Pentagon	Derivation	Infrastructure	Y Combinator profile
flowscope	Derivation	Operating-company	Y Combinator profile
Foaster	Derivation	Operating-company	Y Combinator profile
Perfectly	Derivation	Operating-company	Y Combinator profile
Prism	Derivation	Operating-company	Y Combinator profile
Peer	Derivation	Operating-company	Y Combinator profile
item	Derivation	Operating-company	Y Combinator profile
Dialogus	Derivation	Operating-company	Y Combinator profile
Wealor	Derivation	Operating-company	Y Combinator profile
Minerva	Derivation	Operating-company	Y Combinator profile
Panta	Derivation	Operating-company	Y Combinator profile
Korso	Derivation	Operating-company	Y Combinator profile
Marker	Derivation	Operating-company	Y Combinator profile
Locke	Hold-out	Hold-out / mixed	Y Combinator profile
Hedge	Hold-out	Hold-out / mixed	Y Combinator profile
Trope	Hold-out	Hold-out / mixed	Y Combinator profile
Terminal Use	Hold-out	Hold-out / mixed	Y Combinator profile

Appendix C — Reusable YC-13+5 Assessment Instrument

Do not sum criterion ratings into a composite score. YC-13+5 v1.0 has not established equal criterion weights, necessary/sufficient thresholds or a validated aggregate measure.

#	Criterion	Strength	Evidence	Notes / source
1	Autonomous execution	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
2	Persistent organisational identity and state	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
3	Agent-to-agent coordination	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
4	Dynamic organisational structure and delegation	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
5	Institutional memory	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
6	Explicit authority and roles	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
7	Governance and human-agent boundaries	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
8	Event-driven activation	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
9	Shared capabilities and tools	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
10	Model/vendor-decoupled AI roles	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
11	Elastic AI workforce	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
12	Company state independent of individual workers	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
13	Real-world execution	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
14	Outcome Feedback Loops	NOT DEMONSTRATED / PARTIAL / STRONG / STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
15	Outcome-Calibrated Autonomy	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
16	Executable Governance	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
17	Organisational Observability, Provenance & Auditability	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED
18	Autonomous Resilience & Recovery	NOT DEMONSTRATED / PARTIAL / STRONG / VERY STRONG	E0 ASSERTION / E1 PUBLICLY DOCUMENTED / E2 IMPLEMENTATION INSPECTED / E3 OPERATIONALLY CORROBORATED / E4 INDEPENDENTLY VERIFIED