AI Startup Classifications: What AI Native, AI Augmented, and AI Adjacent Actually Mean

Understanding AI startup classifications is essential for anyone evaluating the 5,300+ AI funding deals Bot Memo tracks. Not every company claiming to be an “AI company” uses AI the same way — some built their entire product on machine learning, others bolted a chatbot onto a 10-year-old SaaS platform, and a third group sells the GPUs and training data that make everything else possible.

These distinctions matter. They shape how investors evaluate deals, how founders position products, and how analysts track a market where AI washing has drawn SEC enforcement actions and $400,000 in penalties.

Bot Memo tracks 5,300+ AI funding deals across 18 primary verticals and 104 tags using a 9-value classification framework. This article breaks down the 4 primary AI startup classifications, explains how to apply them, and shows where other frameworks fall short.

On this page

• The 4 Core AI Startup Classifications (And Why They Matter for Investment)
• AI Native: Companies That Couldn't Exist Without AI
• AI Augmented: When Traditional Software Gets an AI Upgrade
• AI Adjacent: The Picks-and-Shovels of the AI Economy
• AI Platforms: The Foundation Model Builders (3-Question Litmus Test)
• The Removability Test: How to Classify Any AI Startup in 60 Seconds
• Why Other Frameworks (McKinsey, MIT Sloan) Miss the Investor Signal
• FAQ: AI Startup Classification Questions Answered

The 4 Core AI Startup Classifications (And Why They Matter for Investment)

Most AI classification frameworks sort companies by how much AI they use. Bot Memo’s framework sorts by what role AI plays in the product’s existence — a distinction that maps directly to AI investment categories and risk profiles.

Classification	Definition	Key Signal	Example Companies
AI Native	AI is foundational; product couldn’t exist without it	Remove AI and the product disappears entirely	Perplexity, Harvey, Midjourney
AI Augmented	Existing product enhanced with AI features	Remove AI and core product still functions	Salesforce (Einstein), Adobe (Sensei)
AI Adjacent	Enables AI builders but doesn’t run models in-product	Sells picks and shovels, not gold	Scale AI, CoreWeave
AI Platforms	Builds foundational AI technology (models or chips)	Trains own foundation models or designs custom silicon	OpenAI, Anthropic, Cerebras

Source: Bot Memo AI Classification Framework (9-value system applied to 5,300+ tracked deals)

Five additional secondary classifications — Non-AI, Venture Firm, Acquisition, News Article, and Insufficient Info — act as filters to exclude entries that are not AI startup categories from the dataset.

This framework reflects how capital actually flows. An investor evaluating AI native startups faces different risk-return dynamics than one backing AI infrastructure companies. When comparing AI first vs AI native or AI enabled vs AI native, the classification determines which comparables apply, which multiples make sense, and which competitive moats are real.

AI Native: Companies That Couldn’t Exist Without AI

An AI native company is one where artificial intelligence is not a feature — it is the product. Strip out the AI and nothing remains. No fallback workflow. No manual alternative. The product simply ceases to exist.

Perplexity is the textbook example. Founded in 2022, the AI-powered answer engine raised over $1.5 billion and reached a $20 billion valuation by September 2025. Without its large language models, Perplexity has no product. There is no “manual search” mode to fall back on.

Harvey follows the same pattern in legal AI. The company raised $160 million in its Series F at an $8 billion valuation in December 2025, with investors including Andreessen Horowitz and Sequoia. Harvey’s legal AI platform is AI from the ground up — not a document search tool with an AI layer added later.

What separates AI native from AI first?

The terms “AI first” and “AI native” are often used interchangeably, but there is a meaningful difference when distinguishing AI first vs AI native. AI first describes a strategy — a decision to prioritize AI in product development. AI native describes an architecture — the product’s fundamental dependency on AI to function at all.

A company can be AI first (prioritizing AI features in its roadmap) without being AI native (the core product could still work without AI). Every AI native company is AI first by definition. Not every AI first company is AI native. Understanding this AI first vs AI native distinction matters for accurate AI company types classification.

AI native startups typically share three traits:

• Founded during or after the transformer era (2017+, most commonly 2021+)
• No pre-AI product version exists — the company launched with AI as the core
• Data flywheel dependency — the product improves as users interact with it, feeding training data back into the model to create compounding performance gains

For a deeper breakdown of these and related terms, see our AI startup glossary.

AI Augmented: When Traditional Software Gets an AI Upgrade

An AI augmented company is an established product that added AI capabilities to enhance what already existed. The AI improves the experience, but the core value proposition predates it. Understanding AI enabled vs AI native distinctions helps clarify why this category exists separately.

Salesforce introduced Einstein AI in 2016 — 17 years after the company was founded. Einstein adds predictive lead scoring, automated data entry, and conversational insights. Remove Einstein entirely, and Salesforce remains a fully functional CRM.

Deals still close. Pipelines still move. The AI makes things faster and smarter, but the product stands on its own.

Adobe follows the same pattern with Sensei, its AI and machine learning framework. Photoshop’s generative fill and neural filters are impressive. But Photoshop shipped its first version in 1990 — 26 years before Sensei was introduced.

How AI augmented differs from AI enabled

The distinction between “AI augmented” and “AI enabled” is subtle but real. AI enabled typically implies that AI has been switched on — a binary state. AI augmented implies that AI enhances and extends existing capabilities, suggesting a spectrum rather than a toggle. When evaluating AI enabled vs AI native, the augmented category sits in between.

In Bot Memo’s classification framework, companies fall into AI augmented when:

• The company existed and had revenue before adding AI features
• Removing all AI/ML would leave a functional (if less competitive) product
• AI is a differentiator, not the foundation

This matters for AI investment categories because AI augmented companies carry less technical risk (the core business is proven) but face competitive pressure from AI native entrants building the same category from scratch.

AI Adjacent: The Picks-and-Shovels of the AI Economy

AI adjacent companies enable the AI ecosystem without running AI models in their own products. They sell the infrastructure, data, and tooling that AI native and AI platform companies depend on.

Scale AI is the canonical example. The company provides data labeling, data curation, and evaluation services that foundation model companies need to train their models. In June 2025, Scale AI raised $14.3 billion — a record round — bringing its valuation to $29 billion. Yet Scale AI’s own product does not run foundation models. It prepares the data that other companies feed into theirs.

CoreWeave provides another angle on AI adjacent as an AI infrastructure company. The company operates specialized GPU cloud infrastructure — it went public in March 2025 and guided for $12-13 billion in 2026 revenue. CoreWeave does not train models or ship AI-powered products. It rents the compute that model builders need.

The picks-and-shovels advantage

AI adjacent companies often carry lower technical risk than AI native startups. Their revenue is less dependent on model performance breakthroughs. When a new foundation model architecture emerges, AI adjacent companies can pivot to serve it. AI native companies built on the old architecture may not survive the transition.

The tradeoff: AI adjacent companies can become commoditized. GPU clouds compete on price. Data labeling services face pressure from synthetic data. The moat depends on scale, switching costs, and specialization — not proprietary model performance.

AI Platforms: The Foundation Model Builders (3-Question Litmus Test)

AI platforms represent the most capital-intensive and technically ambitious AI startup categories. These are foundation model companies that build foundational AI technology — training their own models from scratch or designing custom silicon for AI workloads.

Bot Memo applies a strict 3-question litmus test. A company qualifies as an AI platform if it answers YES to at least one:

1. Does this company train its OWN foundation models from scratch? (Not fine-tune. Not host. Train from scratch on proprietary data and compute.)
2. Does this company design custom silicon/chips specifically for AI workloads? (Not buy GPUs. Design the chips themselves.)
3. Does this company operate GPU clusters sold specifically as AI training/inference compute? (Dedicated AI compute, not general cloud.)

OpenAI and Anthropic pass on question 1 — they train GPT and Claude models from scratch. Cerebras passes on question 2 — it designs wafer-scale AI chips and raised $1 billion at a $23 billion valuation in February 2026. NVIDIA passes on question 2 as well — its H100 and B200 GPUs are purpose-built for AI training.

What is NOT an AI platform?

The litmus test excludes several companies that are commonly mislabeled:

Company	Why NOT an AI Platform	Correct Classification
Hugging Face	Hosts models, doesn’t train foundation models from scratch	AI Adjacent
LangChain	Orchestration framework, no model training	AI Adjacent
Weights & Biases	MLOps tooling, no model training	AI Adjacent
Databricks	Primarily a data/analytics platform; trained DBRX via Mosaic AI but primary revenue is data lakehouse, not model building	AI Augmented

Source: Bot Memo AI Classification Framework — 3-Question Litmus Test

The distinction between an AI platform company and an AI native company is structural. AI native companies use foundation models as a core dependency. AI platform companies build the foundation models that AI native companies depend on.

The Removability Test: How to Classify Any AI Startup in 60 Seconds

Every AI classification framework needs a practical heuristic. Bot Memo’s is the Removability Test — a single question that maps any company to one of the four primary AI company types:

“If you removed all AI and machine learning from this product, would it still function?”

The answer maps directly to a classification:

• “No — the product disappears entirely.” → AI Native
• “Yes — core product works, but it loses key features.” → AI Augmented
• “The product doesn’t use AI in-product — it enables others who do.” → AI Adjacent
• “No — AND the company trains its own models or designs AI chips.” → AI Platforms

Applying the test: 4 worked examples

Midjourney — Remove the diffusion models and there is nothing left. No manual image editor. No template library. The product is the AI. Classification: AI Native.

Canva — Remove Magic Design, background remover, and text-to-image features. Canva still works as a drag-and-drop design tool with templates, elements, and collaboration. Classification: AI Augmented.

Scale AI — Scale’s core product is data labeling and curation for AI companies. It enables AI builders. Its own product doesn’t run foundation models for end users. Classification: AI Adjacent.

Anthropic — Remove Claude and Anthropic has no product. But Anthropic also trains Claude from scratch on its own infrastructure. This pushes it beyond AI native into AI Platforms. Classification: AI Platforms.

The test takes 60 seconds, works on any company, and produces consistent results across analysts. That is why Bot Memo uses it as the first-pass filter before deeper classification with dual WebSearch verification.

Why Other Frameworks (McKinsey, MIT Sloan) Miss the Investor Signal

Several respected institutions publish their own AI company classification frameworks. Each serves a purpose. None maps cleanly to how capital actually flows.

MIT Sloan’s “Six Types of AI Startups” categorizes companies as originators, explorers, infrastructure builders, enhancers, optimizers, and experimenters. The framework is useful for academic analysis but creates overlapping categories. A company can be both an “originator” and an “infrastructure builder,” which makes the classification ambiguous for investment screening.

McKinsey’s “Takers, Shapers, Makers” framework sorts companies by how they consume AI. Takers use off-the-shelf models, shapers customize them, and makers build from scratch. This is a useful maturity model but tells investors nothing about the product’s dependency on AI.

A “taker” that built its entire product on GPT-4 API calls is AI native — but McKinsey’s framework puts it in the same bucket as a company that added ChatGPT to its help desk.

Sapphire Ventures’ 5D framework evaluates AI-native applications across multiple dimensions including UI paradigm, data network effects, and pricing model. This framework is strong for evaluating AI-native quality but doesn’t classify companies that aren’t AI native. It answers “how good is this AI-native company?” not “what type of AI company is this?”

Bot Memo’s AI classification framework addresses the investor signal gap: it answers a binary structural question (what role does AI play?) before moving to qualitative evaluation (how good is the AI?). This maps directly to investment thesis categories, comparable company sets, and risk profiles.

How the 9-value classification pipeline works

Bot Memo’s classification pipeline uses 3 parallel AI classifiers, each performing dual WebSearch verification per company. A company only receives its final classification when at least 2 of 3 classifiers agree. Disagreements trigger manual review.

The 9-value system (4 primary + 5 secondary filter values) processes every company entering Bot Memo’s dataset of 5,300+ tracked deals across 18 primary verticals and 104 tags. For details on what happens after classification, see Bot Memo’s 18 primary verticals and 104 tags.

This matters because AI washing — where companies exaggerate their AI capabilities — remains a persistent problem. The SEC created a dedicated Cyber and Emerging Technologies Unit (CETU) in February 2025 specifically to pursue AI washing enforcement. Bot Memo’s dual-verification approach is designed to catch companies that claim AI native status but fail the removability test. For more on identifying this pattern, see our guide on how to spot AI washing.

FAQ: AI Startup Classification Questions Answered

What is the difference between AI native and AI augmented?

An AI native company built its product on AI from the ground up — without AI, the product does not exist. An AI augmented company added AI features to an existing product that was already functional. Perplexity (AI-powered answer engine, founded 2022) is AI native. Salesforce (CRM since 1999, added Einstein AI in 2016) is AI augmented. The Removability Test makes this concrete: remove all AI from Perplexity and nothing remains. Remove Einstein from Salesforce and you still have a working CRM.

How do investors classify AI startups?

Most venture capital firms use informal classification during deal screening, sorting companies by how central AI is to the product. Bot Memo formalizes this into 4 primary AI investment categories — AI Native, AI Augmented, AI Adjacent, and AI Platforms — each carrying different risk-return profiles. AI Platforms require the most capital (foundation model training costs exceed $100 million per frontier run). AI Adjacent companies carry the lowest technical risk. AI Native startups offer the highest potential returns but depend on continued AI capability improvements.

What does AI adjacent mean for startups?

AI adjacent describes companies that enable the AI ecosystem without running AI models in their own products. These are the picks-and-shovels businesses:

• Data labeling (Scale AI)
• GPU cloud providers (CoreWeave)
• MLOps platforms (Weights & Biases)
• Model hosting services (Hugging Face)

AI adjacent startups benefit from AI growth regardless of which models or approaches win, but face commoditization pressure as the infrastructure layer matures.

What is an AI platform company vs an AI native company?

Both depend entirely on AI — remove AI and neither product survives. The difference is structural. An AI platform company builds foundational AI technology: it trains its own foundation models from scratch (OpenAI, Anthropic, Mistral) or designs custom AI chips (Cerebras, NVIDIA). An AI native company uses foundation models as a dependency but does not build them. Harvey is AI native — it built a legal AI product on top of large language models. Anthropic is an AI platform — it trains Claude from scratch. Bot Memo’s 3-question litmus test distinguishes between the two.

How do you tell if a company is truly AI native?

Apply the Removability Test: strip out all AI and ML components. If the product vanishes entirely, it is AI native. If a functional (if diminished) product remains, it is AI augmented. Beyond this binary test, look for three signals: (1) the company was founded during or after the transformer era with no pre-AI product version, (2) AI appears in the core product loop, not just a feature sidebar, and (3) the company’s data flywheel — where user interactions improve model performance — is central to the value proposition, not an afterthought.

What are the different types of AI companies?

AI company types generally fall into four categories based on what role AI plays in the product. AI Native companies (like Perplexity or Harvey) could not exist without AI. AI Augmented companies (like Salesforce or Adobe) added AI features to an already-functional product. AI Adjacent companies (like Scale AI or CoreWeave) sell infrastructure and tooling to AI builders without running AI models themselves. AI Platforms (like OpenAI or Anthropic) build foundational technology — training models from scratch or designing custom AI chips. Bot Memo’s 9-value classification framework applies these four primary categories plus five secondary filters to every company in its dataset.

What is AI washing and how does it relate to classification?

AI washing occurs when companies exaggerate or fabricate their use of artificial intelligence to attract investors, customers, or media attention. The SEC has taken enforcement action against AI washing, levying $400,000 in penalties, and its Cyber and Emerging Technologies Unit (CETU) specifically targets misleading AI claims. Rigorous AI startup classifications are the primary defense against AI washing. Bot Memo’s framework uses the Removability Test and dual WebSearch verification through 3 parallel classifiers to catch companies that claim AI native status but whose products would function without AI — a hallmark of AI washing.

---

Bot Memo tracks 5,300+ AI funding deals across 18 primary verticals and 104 tags. Our AI classification framework assigns one of 9 values to every company entering the dataset, verified through dual WebSearch queries and 3 parallel classifiers. Explore the full taxonomy at Bot Memo’s 18 primary verticals and 104 tags.