Discontinuous Bet Incubation
Pursue a bold product theme through small, disposable experiments.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 7
- Confidence
- 96%
Start by identifying a strategically important theme that could produce a discontinuous outcome rather than an incremental feature. Define what the idea might look like at 10x, 100x, or 1000x scale, but avoid becoming attached to the first implementation. Give a small, autonomous pod—or even one founder-like engineer—room to build prototypes and gather evidence quickly. Evaluate whether there is genuine user value, whether the technology is ready, and what the experiment teaches about the opportunity. Bets that are not working should be stopped without treating the effort as wasted, because a prototype can produce valuable organizational learning. Preserve the thesis and revisit it after one or two technology generations if the constraint was capability rather than demand. This structure combines strategic conviction with tactical flexibility.
Origin
Extracted from Lenny's Podcast, where Dianne Penn explained the operating thesis behind Anthropic's Labs organization and products including Claude Code, skills, MCP, and Claude Design.
Core principles
- 01Seek opportunities outside the incremental core roadmap.
- 02Hold the strategic theme strongly and the current prototype weakly.
- 03Ask what the 10x, 100x, or 1000x version could become.
- 04Use small founder-like teams for ambiguous bets.
- 05Treat learning as a valid result even when nothing ships.
- 06Revisit promising ideas when the enabling technology improves.
How to run it
- 1
Choose a discontinuous theme
Identify an opportunity that could materially change user behavior or create a new product category. Keep it separate from routine core-roadmap optimization.
- 2
Expand the ambition
Describe the 10x, 100x, and 1000x versions of the opportunity. Use these horizons to reveal a larger destination than the initial prototype.
- 3
Form a tiny pod
Give a small team or individual strong ownership of the experiment. Select people who enjoy zero-to-one ambiguity and can operate like founders.
- 4
Prototype loosely
Build a concrete experiment while remaining willing to replace the exact interface, workflow, or technical approach. Optimize for learning rather than polish.
- 5
Judge the evidence
Assess whether users experience meaningful value, whether the technology is ready, and whether the opportunity could grow into the ambitious vision.
- 6
Stop or scale
Turn off experiments that lack sufficient evidence, or expand those that reveal a credible there-there. Move validated bets toward a durable product vehicle.
- 7
Revisit with better technology
Archive promising theses whose current blocker is model capability and reconsider them after one or two model generations.
In the wild
Anthropic treated agentic coding as a discontinuous opportunity and developed Claude Code as a product vehicle rather than relying on model intelligence alone. A later frontier model, Opus 4.5, could then express its capabilities through an end-to-end agentic experience, while the stronger model accelerated adoption of the product.
→ The combination of a frontier product and frontier model created an inflection that neither component was likely to achieve independently.
A one-engineer pod tests an agent that can complete a complex professional workflow. Users like the concept, but the current model fails too often at a critical reasoning step. The team stops active development, records the capability threshold it needs, and retests after two model generations rather than scaling an unreliable product.
→ The organization preserves the insight while avoiding a large premature investment.
Common mistakes
Staffing ambiguity with a large team
A large team creates communication and coordination costs before the opportunity is understood. Begin with the smallest credible ownership unit.
Falling in love with the prototype
The strategic area may be correct even when the current implementation fails. Change the prototype without abandoning the underlying thesis prematurely.
Treating stopped bets as failures
A prototype that identifies a missing capability or invalid assumption can still create reusable knowledge. Preserve that learning and its revisit conditions.
Is it for you?
Best for
It is best for frontier-technology teams facing many plausible products, uncertain timing, and rapidly changing technical capabilities.
Not ideal for
It is not ideal for routine optimization work with stable requirements, predictable technology, and clear delivery commitments.
From the transcript
“The thesis of labs in many ways is identifying and pulling the thread on the thread of discontinuous large bets that might not be in…”
“you can be very strongly held opinion about the theme or the area, and then more weakly held about the exact prototype.”
“sometimes these ideas start with one engineer, right?”
From the episode
Anthropic’s first technical PM on token maxing, the jagged edge, and living in the future