The short version
What "AI-native" actually means
"AI-native" has moved from telecom standards documents into boardroom language. Usage is loose, but the serious definitions agree on one test: is AI an add-on, or is it the foundation the business is built on?
An AI-native company is an organization whose products, operations and decision-making are designed around AI from the ground up. Data and learning loops are core assets, AI agents and models do a meaningful share of the work, and humans focus on direction, judgement and accountability.
Three published framings are useful:
Intrinsic, trustworthy AI
"AI native is the concept of having intrinsic trustworthy AI capabilities, where AI is a natural part of the functionality, in terms of design, deployment, operation, and maintenance." It also replaces "static, rule-based mechanisms with learning and adaptive AI."[1]
The agentic organization
McKinsey calls this "the largest organizational paradigm shift since the industrial and digital revolutions." Work is rebuilt around "AI-first workflows", with humans reintroduced where they add the most value, and teams are organized as "flat networks of empowered, outcome-aligned agentic teams."[2]
Core business model test
ICONIQ's 2025 survey classes a company as AI-native when its "core product or business model is AI-driven." It separates these from AI-enabled firms that are building new AI products or adding AI features to existing ones.[6]
A simple heuristic from the venture community: if you take the AI out and the product or operating model stops working, the company is AI-native. If it still works, only less efficiently, the company is AI-enabled.
Digital-native, AI-enabled and AI-native
The clearest way to understand AI-native is to compare it with what came before. Digital-native companies (born on the web and cloud) put software at the core. AI-enabled companies add AI tools to existing software and processes. AI-native companies redesign the core around models, data and agents.
| Dimension | Digital-native | AI-enabled | AI-native |
|---|---|---|---|
| Role of AI | Minimal or analytics only | Productivity add-on to existing tools | Foundation of product and operations |
| Product | Deterministic software features | Software with AI features attached | Product capabilities that only exist because of models |
| Data | Stored for reporting | Fragmented and partly accessible to AI | Strategic asset feeding continuous learning loops |
| Workflows | Digitized manual steps | Same steps, assisted by AI | Redesigned end-to-end around agents and human checkpoints |
| Team shape | Scales headcount with growth | Similar headcount, some efficiency gains | Small, senior teams orchestrating AI. High output per person |
| Decision-making | Dashboards and intuition | AI-generated insights, human-run | AI recommends and executes within policy. Humans own the critical calls |
| Governance | IT and security policies | Ad-hoc AI usage policies | Built-in permissions, evaluation, audit and risk controls |
| Pricing | Per seat or subscription | Per seat plus AI add-on | Increasingly usage- or outcome-based |
Eight traits of AI-native companies
Across analyst research, investor studies and practitioner accounts, the same traits keep appearing. No company shows all of them perfectly, but together they describe the target state.
AI at the core of value
The core product or service depends on models. Take AI away and the value proposition disappears, rather than just slowing down.
Data and learning loops as assets
Every interaction produces signals (corrections, outcomes, preferences) that are captured and used to improve models, prompts and context.
Agentic workflows
Processes are redesigned so AI agents plan and execute multi-step work using tools and systems. They escalate to people at defined checkpoints.
Small, high-leverage teams
Revenue and output per employee are far above traditional benchmarks. Hiring is focused on judgement, taste and system design.[7]
Governance by design
Permissions, human approval gates, evaluation and audit trails are built into the architecture from the start, not added after an incident.
Rapid experimentation
Short build-measure-learn cycles, with evals instead of opinions. Models and vendors are swapped as the frontier moves.
Knowledge made machine-readable
Policies, runbooks, decisions and domain expertise are structured so AI can retrieve and act on them with the right permissions.
The architecture of an AI-native organization
Being AI-native is above all an architectural decision. The diagram below is a vendor-neutral reference model built from common patterns in enterprise AI platforms.[12][1] It has six horizontal layers, which carry work from infrastructure up to the people who use it, and two vertical layers that run through every level: governance and continuous evaluation.
Foundation
Reliable cloud infrastructure and a unified, well-governed data platform. Most AI programmes that stall do so here, on data quality, access and integration.
Intelligence
Company knowledge made retrievable and permission-aware, served by a mix of models behind a gateway, so the organization is never locked into one vendor.
Orchestration
Agents and role-specific assistants do work across systems, with human approval gates and full traceability, delivered where people already work.
How a request flows through the stack
The architecture is easier to understand by following one task. Here an engineering lead asks an assistant to prepare a cloud cutover plan:
Why AI-native companies compound
The structural advantage of an AI-native company is a learning loop. More usage produces more proprietary signals: corrections, outcomes, edge cases and domain context. Those signals improve the system, which drives more usage. Competitors can buy the same foundation models. They cannot buy your loop.
How the organization itself changes
Technology is half the story. AI-native companies are organized differently. The traditional pyramid, with many people executing tasks under layers of coordination, gives way to smaller teams of senior people who each direct a set of AI agents. Each person's span of influence grows. Work shifts from doing to specifying, reviewing and deciding.
New roles appear
Agent designers, AI product owners, evaluation engineers and knowledge stewards. Domain experts become reviewers and trainers of AI behaviour.
Judgement over throughput
The scarce skills are problem framing, critical review, systems thinking and accountability. Typing speed and rote execution matter much less.
New measures of performance
Cost per task, cycle time, automation rate, human override rate, eval scores and revenue per employee sit beside traditional KPIs.
What the numbers say
AI adoption is now nearly universal. Value is not. The evidence shows a widening gap between the few organizations built for AI and the majority that have bolted it on.
Agents are the next wave, and the next risk
33% of enterprise apps
will include agentic AI, up from under 1% in 2024. At least 15% of day-to-day work decisions will be made autonomously.[9]
40%+ of agentic projects cancelled
because of rising costs, unclear business value or inadequate risk controls. Gartner also warns of widespread "agent washing."[9]
~40% of AI value from agents
is expected by 2030. 42% of companies expect to give agents genuine autonomy, but only 5% have the full set of controls in place.[4]
~80% of AI-native builders
are investing in agentic systems that take multi-step actions for users. AI-native products use an average of 2.8 models each.[6]
AI-native in practice
AI-native companies come in two forms. Some were born AI-native, founded around models from day one. Others are transforming incumbents rebuilding their operating model. Revenue figures for private companies come from press reports and are approximate.
Born AI-native
Cursor (Anysphere)
Software devAn AI-first code editor where agents write, refactor and review code alongside developers.
Glean
Enterprise knowledgeA permission-aware enterprise search and agent platform built on a company's own knowledge graph.
ElevenLabs
Voice AIVoice models and agents for media, customer service and accessibility. The model is the product.
Harvey
LegalA domain-specialised AI platform for law firms, an example of "AI-native services" in a professional vertical.
Midjourney
Generative mediaImage generation built and scaled with no external funding and a very small team.
Intercom Fin
Outcome pricingAn AI customer-service agent priced per resolved conversation instead of per seat. This is the AI-native business model in practice.
Transforming incumbents
Shopify
Operating ruleThe CEO's 2025 memo made "reflexive AI usage" a baseline expectation. Teams must show why AI cannot do the work before asking for headcount, and AI use counts in performance reviews.[19]
Duolingo
AI-first memoDeclared itself "AI-first" in April 2025. It phased out contractors for work AI can do and built AI proficiency into hiring and reviews. The memo drew public backlash, a reminder that change management matters.[20]
Moderna
Org redesignRolled out ChatGPT Enterprise company-wide with hundreds of custom GPTs. It then merged HR and technology under one leader, treating workforce and AI design as a single problem.[22]
Klarna
Cautionary caseIts AI assistant handled two-thirds of customer-service chats in its first month (2024). By 2025 the CEO admitted the focus on cost had hurt quality and resumed hiring human agents. AI-native is not the same as human-free.[21]
Five stages from experimenting to AI-native
Organizations rarely jump straight to AI-native. Most move through recognisable stages. Each stage needs specific foundations before the next can be reached safely. The model below combines the stages used across analyst maturity frameworks (Gartner, Ericsson, BCG) into a practical sequence.[11][1][5]
| Stage | Typical signs | Foundation needed to advance | Main risk |
|---|---|---|---|
| 1 · Experimenting | Scattered pilots, personal ChatGPT use | AI policy, use-case inventory, executive sponsor | Data leakage via shadow AI |
| 2 · Adopting | Licensed copilots, training programmes | Data access, identity and permissions, success metrics | Usage without measurable value |
| 3 · Integrating | AI in specific workflows, retrieval over company docs | Unified knowledge layer, evaluation, integration platform | "Pilot purgatory" with no scale |
| 4 · Orchestrating | Role-based agents taking actions with approvals | Orchestration layer, governance plane, LLMOps | Uncontrolled agent actions |
| 5 · AI-native | Workflows redesigned, new products and pricing | Org redesign, new roles, continuous learning culture | Over-automation and loss of judgement |
What can go wrong, and how to control it
Becoming AI-native increases both reward and exposure. The companies that succeed treat governance as an enabler of speed rather than a brake: clear rules let teams move faster with confidence.
The value gap
Most organizations struggle to turn adoption into profit. Only about a third see EBIT impact. MIT NANDA found that tools which do not learn from feedback are a leading cause of failure.[3][10]
Control: tie every use case to a measured business outcome, and build feedback loops in from day one.Uncontrolled agents
Prompt injection is the top risk in the OWASP Top 10 for LLMs, and "excessive agency" has grown in importance as systems take actions.[13]
Control: least-privilege permissions, sandboxed tools, approval gates for consequential actions.Hallucination & quality drift
Models can be confidently wrong, and quality can shift silently when models or data change. Klarna shows how customer experience can suffer.[21]
Control: grounded retrieval, continuous evals, human review where errors are costly.Cost & margins
Inference is a real cost of goods. Bessemer's fastest-growing AI companies run ~25% gross margins, against ~60% for conventional peers.[7]
Control: model routing, caching, small models for routine tasks, cost-per-task tracking.Regulation
The EU AI Act uses a risk-based approach. Requirements and application dates depend on the system, its use and the organization’s role. Use the European Commission’s current guidance when mapping obligations.[23]
Control: an AI system inventory, risk classification and documentation from the start.A practical path for established organizations
Startups can be born AI-native. Established organizations have to become AI-native while keeping the business running. The research points to a staged approach: build the foundations, prove value in focused workflows, then scale through a governed orchestration layer.
Understand
Clarify business goals and constraints. Map where knowledge work happens, where decisions bottleneck and where AI already appears (including shadow AI).
Assess
Assess AI readiness across data, architecture, security, skills and governance. Rank use cases by value, feasibility and risk.
Design
Define the target-state architecture (knowledge, model, orchestration and governance layers) and a roadmap with measurable outcomes for each stage.
Enable
Build the knowledge and orchestration foundations. Deploy role-based assistants with human-in-the-loop controls, and train people on the new ways of working.
Evolve
Run evaluations continuously, monitor cost and quality, expand automation where trust has been earned, and redesign workflows end-to-end.
Where is your organization on the AI-native journey?
Moderta Native helps organizations move from isolated AI tools to a governed, AI-native operating capability, from readiness assessment and architecture to role-based assistants with human approval built in.
Book a consultation →References
- Ericsson, "Defining AI native" (white paper). https://www.ericsson.com/en/reports-and-papers/white-papers/ai-native
- McKinsey, "The agentic organization: Contours of the next paradigm for the AI era". https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-agentic-organization-contours-of-the-next-paradigm-for-the-ai-era
- McKinsey, "The state of AI" (2025). https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-2025
- BCG, Applied AI Index press release (30 Sep 2026). https://www.bcg.com/press/30september2026-ai-starting-to-pay-off-companies-generate-value
- BCG, "The Widening AI Value Gap" press release (30 Sep 2025). https://www.bcg.com/press/30september2025-ai-leaders-outpace-laggards-revenue-growth-cost-savings
- ICONIQ, "2025 State of AI: The Builder's Playbook". https://www.iconiq.com/growth/reports/2025-state-of-ai
- Bessemer Venture Partners, "The State of AI 2025". https://www.bvp.com/atlas/the-state-of-ai-2025
- Menlo Ventures, "2025: The State of Generative AI in the Enterprise". https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/
- Gartner, AI agents research and predictions (2025); coverage via Allwork.Space. https://www.gartner.com/en/articles/ai-agents-pc1
- MIT NANDA, "The GenAI Divide" (2025), via Virtualization Review (preliminary study). https://virtualizationreview.com/articles/2025/08/19/mit-report-finds-most-ai-business-investments-fail-reveals-genai-divide.aspx
- Overview of AI maturity models incl. Gartner (BMC). https://www.bmc.com/resources/topics/ai-maturity-models.html
- a16z, "Emerging Architectures for LLM Applications". https://a16z.com/emerging-architectures-for-llm-applications/
- OWASP Top 10 for LLM Applications 2025 (summary, Invicti). https://www.invicti.com/blog/web-security/owasp-top-10-risks-llm-security-2025
- Cursor Series D coverage (Crowdfund Insider, Nov 2025). https://www.crowdfundinsider.com/2025/11/255630-ai-coding-startup-cursor-raises-2-3bn-valued-at-29-3bn-in-series-d/
- Glean press release, $200M ARR. https://glean.com/press/glean-surpasses-200m-in-arr-for-enterprise-ai-doubling-revenue-in-nine-months
- ElevenLabs ARR coverage (CXO Digital Pulse). https://www.cxodigitalpulse.com/?p=53213
- Harvey ARR and valuation coverage (reported). https://entrepreneurloop.com/harvey-ai-valuation-15-5b/
- Sacra, Midjourney revenue estimates. https://sacra.com/research/midjourney
- Shopify CEO AI memo coverage (Windows Central). https://www.windowscentral.com/software-apps/work-productivity/shopify-ceo-lists-ai-skills-as-fundamental-expectation
- Duolingo AI-first memo coverage (Computing). https://www.computing.co.uk/news/2025/ai/duolingo-goes-ai-first-as-it-phases-out-human-contractors
- Klarna reverses course on AI customer service (Entrepreneur). https://www.entrepreneur.com/business-news/klarna-ceo-reverses-course-by-hiring-more-humans-not-ai/491396
- OpenAI, Moderna customer story; Dataconomy on HR/tech merger. https://openai.com/index/moderna
- European Commission, AI Act: risk categories, obligations and application timeline. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
- Intercom pricing (Fin). https://www.intercom.com/pricing
Note: Revenue and valuation figures for private companies come from press reports and estimates and may change. Statistics are dated findings from the cited studies, not guarantees of future performance. Diagrams are original illustrations by Moderta Native.
