When Foundation Models Commoditize Your AI Product: The Commoditization Risk Strategy
TL;DR
Every quarter, OpenAI and Anthropic ship features that commoditize an entire product category. Jasper got commoditized by ChatGPT. Summary apps got commoditized by Apple Intelligence. Generic image-prompt tools got commoditized by ChatGPT image generation. The question isn't whether it happens to you — it's how you've designed your product to survive when it does. There are three real defenses (workflow lock-in, data accumulation, vertical depth), ranked from weakest to strongest. Pricing strategy is the absorber. The "feature vs product" test tells you in 60 seconds whether you're at risk this quarter.
The Commoditization Clock: Which Categories Die When
Commoditization isn't random. Foundation model providers ship in a predictable pattern: the categories they can absorb cheapest, with the highest visibility, go first. The categories that require deep integration, regulated workflows, or proprietary data are last — or never. If you map your product to where it sits on this clock, you know how much time you have.
0-6 months (Already commoditized)
Prompt-only wrappers, basic chat UIs, prompt marketplaces, simple summarization tools, generic copywriting. These have been absorbed by ChatGPT, Claude, and Apple Intelligence. If this describes your product, the commoditization already happened — your defense is wedge migration, not survival.
6-18 months (Currently commoditizing)
Generic meeting transcription, basic image generation, single-step coding assistants, document Q&A without enterprise integrations, mid-tier translation. ChatGPT Voice, Apple's native APIs, and the latest GPT/Claude features are eating these now. Plan exit or wedge deepening urgently.
18-36 months (At risk)
Agent-based workflows with weak vertical specificity, RAG-over-public-data products, single-vertical tools without deep customer data, generalist analytics co-pilots. Foundation models are building agents and connectors that will encroach. You have 1-2 cycles to deepen your moat.
36+ months (Protected for now)
Vertical AI with regulated data, products with strong workflow lock-in and per-customer fine-tuning, agents that act on proprietary enterprise systems, anything requiring SOC 2 + HIPAA + per-tenant isolation. Foundation models won't enter these because the unit economics and risk profile don't fit.
The honest exercise: place your product on this clock. Most AI PMs are 6-12 months more optimistic than reality. Talk to engineers at OpenAI or Anthropic informally (or read their public roadmaps) and the commoditization timeline tightens. For a related framing on durable advantages, see AI competitive moats.
Three Defenses, Ranked Weakest to Strongest
There are exactly three durable defenses against foundation-model commoditization. We rank them in increasing order of durability. If your strategy doesn't include at least one (and ideally two), you're betting on luck.
Defense 1: Workflow Lock-In (Weakest)
What happens: Embed so deeply in the user's daily workflow that swapping you out means re-learning a job. Notion AI is inside the document tool people already use. Cursor is the IDE developers live in. The model can be ChatGPT-equivalent — the wrapper has the workflow.
PM Implication: This is the most accessible defense but also the most fragile. Switching costs erode as foundation models add their own workflow features (ChatGPT Canvas, Claude Projects). Buys you 12-24 months — use it to build defense 2 or 3 underneath.
Defense 2: Data Accumulation (Stronger)
What happens: Your product generates proprietary training data that improves your AI faster than competitors. Decagon improves on each customer's resolved tickets. Glean improves on each enterprise's permission and search data. The data exhaust is the moat — and it compounds.
PM Implication: This is durable only if (a) the data is genuinely proprietary, (b) it materially improves your AI's performance, and (c) competitors can't access equivalent data. Most claimed 'data moats' fail one of these three. Be honest in self-evaluation.
Defense 3: Vertical Depth (Strongest)
What happens: Build for a specific vertical with workflows, terminology, integrations, and compliance that foundation models won't replicate because the TAM doesn't justify it. Harvey for legal, Hippocratic for healthcare, Abridge for clinical notes. The depth is the durable advantage.
PM Implication: Strongest defense because it's structurally unattractive for foundation models to enter. The risk: TAM ceiling is lower than horizontal plays. Accept it. A defensible $200M ARR in a vertical beats a vulnerable $50M ARR in a horizontal category that disappears next quarter.
The best AI products stack defenses. Cursor has workflow lock-in (defense 1) plus emerging data accumulation from how developers interact with its agent (defense 2). Harvey has vertical depth (defense 3) plus workflow lock-in (defense 1) plus accumulated training data on legal-specific reasoning (defense 2). Three-layer defenses are how you get to "uncommoditizable."
Pricing Strategy as Commoditization Absorber
Pricing isn't just monetization — it's a shock absorber for commoditization. When a foundation model commoditizes your core feature, the right pricing model gives you headroom to migrate users to higher-value features. The wrong pricing model collapses revenue overnight.
Per-seat pricing absorbs commoditization
If you charge $30/user/month for a platform, individual feature commoditization doesn't kill you — you keep the seat. Notion at $10-$25/user weathered ChatGPT's encroachment because the platform value held even as individual AI features got cheaper to provide.
Per-usage pricing exposes you
If you charge $0.50 per generated image and Midjourney drops to $0.05 effective, you have hours, not quarters. Per-usage works only when paired with depth — Midjourney itself can use it because they're vertically deepest.
Outcome-based pricing is the future
Charging per resolved ticket, per closed deal, per filed compliance form — these don't collapse when foundation models get cheaper, because the value is the outcome. Decagon charges per resolved customer-service interaction. Sierra is moving the same direction.
Bundling protects vulnerable features
If your AI feature is commodity-adjacent, bundle it inside a broader product. Adobe Firefly inside Creative Cloud. Microsoft Copilot inside Office. Customers don't pay for the AI line item — they pay for the platform, which includes AI.
The pricing migration we recommend for products at commoditization risk: shift from per-usage to per-seat over 12-18 months, then layer outcome-based pricing for premium tiers. This three-step migration is the same pattern we walk through in AI product differentiation.
Build Commoditization-Proof Products
The masterclass walks AI PMs through commoditization risk mapping, defense layering, and pricing migration — using their own product roadmap as the case study.
The "Feature vs Product" Test
The 60-second diagnostic: can your entire product be described as a feature inside another product? If yes, you're at imminent commoditization risk. If no, you may still have time. Run this test on yourself before someone with a $100B war chest runs it on you.
Test 1: One-sentence description
If you can describe your product in one sentence that starts with 'It's like ChatGPT but for...', you're a feature. 'AI meeting summarizer' is a feature. 'Clinical documentation platform with HIPAA-compliant ambient capture, EMR integration, and physician review workflow' is a product.
Test 2: Substitutability test
If a customer could replicate 70% of your value by writing a 200-word ChatGPT prompt, you're a feature. If replicating you requires building integrations to 12 enterprise systems, training on 5 years of proprietary data, and passing SOC 2 audit, you're a product.
Test 3: Workflow vs task
Features do tasks. Products own workflows. 'Generate a summary' is a task. 'Manage the entire customer-support workflow from incoming message to resolution to QA to coaching the agent who handled it' is a workflow. Foundation models ship tasks, not workflows.
Test 4: Retention durability
What's your 12-month retention if your core AI feature got commoditized to free tomorrow? If it's above 70%, you have a real product. If it's below 30%, you have a feature with a Stripe integration.
We've watched hundreds of AI product reviews. The teams that take this test seriously and act on a "feature" result aggressively (deepen vertical, add workflow, accumulate data) survive. The teams that score themselves as a "product" without honest evidence get commoditized on schedule. For broader recovery patterns when commoditization hits, see AI failure recovery strategy.
Case Studies: Jasper, Otter, Notion AI, Cursor
Four real cases. Two have been commoditized (with one trying to recover). One absorbed the threat. One has, so far, defended successfully. The differences are instructive.
Jasper — Commoditized by ChatGPT
Jasper raised $125M at a $1.5B valuation in late 2022 on AI copywriting. ChatGPT launched a month later. By 2024, ChatGPT free tier did 80% of Jasper's job. Jasper laid off staff, repositioned as 'enterprise marketing platform.' The lesson: a prompt-only wrapper has no defense against the model provider shipping a UI.
Otter.ai — Being commoditized by Apple Intelligence
Generic meeting transcription is now native on Apple devices and inside Zoom/Teams. Otter is mid-migration to workflow features (CRM integration, sales-specific summaries) — defense 1. Whether that's enough depends on how aggressively Microsoft and Salesforce push their native versions.
Notion AI — Absorbed the threat
Notion didn't try to be the AI tool. They embedded AI inside the workspace people already use. Per-seat pricing absorbed commoditization risk — even when ChatGPT got better, Notion subscribers didn't churn because the platform value held. This is the model for incumbents and rare for startups.
Cursor — Defending successfully (so far)
Cursor has stacked all three defenses: workflow lock-in (IDE), data accumulation (developer interactions improve the agent), and emerging vertical depth (coding-specific UX foundation models won't build). $200M+ ARR by mid-2026, and growing despite Copilot, Claude Code, and Codex all competing.
Pattern across all four
Survival correlates with defense stacking, not feature quality. Jasper had good features. Otter has good features. The question is whether the features sit inside a defensive structure foundation models can't replicate cheaply. The answer determines 18-month survival.
What to do this week
Map your product on the commoditization clock. Score your three defenses honestly. Identify the weakest layer and dedicate Q3 roadmap to deepening it. This isn't optional — it's the AI PM job in 2026. Pair this with the broader defensibility playbook (linked below) for the full picture.
The full 2026 framework — including the migration plans for products mid-commoditization — is in our AI defensibility playbook 2026.