Incumbent AI Feature Absorption
Watch frontier experiments, prototype leanly, then embed proven ideas in core products.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 5
- Confidence
- 90%
Incumbent AI Feature Absorption treats emerging AI applications as a distributed experiment layer. Product teams observe which narrow experiences attract users, identify the durable capability beneath each application, and prototype it inside an established workflow using a lean skunkworks team. Because advanced models can often perform routine tasks with limited prompt engineering, validation and implementation may move faster than conventional software development. Successful features are then integrated into products users already trust and use, giving incumbents advantages in distribution, context, and workflow continuity. The framework predicts that many people will encounter AI through upgraded office, productivity, and business applications rather than permanently switching to numerous standalone AI products.
Origin
Nathan Labenz outlined this application-layer strategy while discussing how existing software companies would respond to fast-moving AI startups on Marketing Against The Grain.
Core principles
- 01Small AI applications reveal useful product experiments.
- 02Established products possess distribution and workflow context.
- 03Reliable model behavior makes prototyping unusually fast.
- 04Most users adopt AI through software they already use.
How to run it
- 1
Observe Frontier Experiments
Track small AI applications and identify which experiences earn sustained user engagement rather than launch-day attention.
Pro tip Focus on behavior and outcomes, not the startup's branding.
- 2
Extract the Durable Capability
Define the underlying task or interaction that creates value and determine whether it belongs in the existing product.
Pro tip Ask whether users would prefer the capability inside their current workflow.
Watch out Do not copy superficial interface details without understanding the user need.
- 3
Prototype Leanly
Give a small team authority to build a narrow integration with the new model.
Pro tip Use direct model performance to reduce unnecessary infrastructure initially.
Watch out Fast prototyping does not eliminate security or data-governance obligations.
- 4
Validate in Context
Test whether the feature improves completion time, output quality, or retention inside the incumbent workflow.
Pro tip Compare it against the user's existing method, not only against other AI demos.
- 5
Integrate or Retire
Productize validated experiments within the core application and discard those that fail to create durable value.
Pro tip Reuse existing distribution and permissions where appropriate.
Watch out Accumulating disconnected AI features can make the core product less coherent.
In the wild
A productivity-suite company observes successful standalone tools for drafting, summarizing, spreadsheet formulas, and presentations. Small teams prototype those behaviors in the applications customers already use, validate them with existing workflows, and roll the best capabilities into the suite.
→ Users gain AI assistance without moving their documents and processes into a separate collection of tools.
Common mistakes
Copying Every Experiment
Many frontier products demonstrate novelty rather than a durable user need.
Moving at Legacy Speed
A large incumbent can lose its distribution advantage if routine AI prototypes remain trapped in normal multi-quarter development cycles.
Ignoring Workflow Context
Adding a model feature without fitting existing user behavior produces an isolated novelty rather than an adopted capability.
Is it for you?
Best for
Incumbent software businesses with existing distribution, user data, and workflows that can benefit from model capabilities.
Not ideal for
Organizations whose architecture, culture, or release process prevents rapid experimentation and integration.
From the transcript
“I think the incumbents ultimately are going to have a lot of advantage in that they are going to be able to see these little…”
“It's something that really lends itself to a kind of lean skunk works, you know, prototyping sort of thing.”
“I think that's kind of what I would expect to happen is that the products that you already use are going to all become AI…”
From the episode
GPT-4 Beta User Reveals What Jobs It Will Destroy In 2023 with Nathan Labenz (#103)
Nathan Labenz