Latent Demand Product Discovery
Find your next product by watching people misuse your current one for something it wasn't designed to do.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 92%
Latent demand is the signal that appears when users hack or repurpose a product to accomplish something it was never built for. Instead of guessing what to build, you watch where people are already jumping through hoops, then build a purpose-built product for that job. Boris extends the classic framing with a modern twist for AI: also watch what the model itself is trying to do and make that easier.
Origin
Boris Cherny credits the classic framing to Fiona, the founding manager of Facebook Marketplace, who observed that ~40% of Facebook group posts were people buying and selling. The AI extension ('look at what the model is trying to do') is Boris's own framing from building Claude Code and Cowork at Anthropic.
Core principles
- 01The strongest product signal is people using an existing product in a way it wasn't designed for
- 02'Abuse' here means creative misuse, not security abuse — it reveals a real unmet job
- 03Bring the product to where people already are rather than forcing a new workflow
- 04In AI, latent demand also means observing what the model naturally reaches for and reducing friction on that path
How to run it
- 1
Instrument for surprising usage
Watch behavioral signals — what portion of activity is people doing something the product wasn't built for. Facebook noticed 40% of group posts were buying/selling; Anthropic noticed people using Claude Code for non-coding tasks.
Pro tip Quantify it. A percentage ('40% of posts', '60% of profile views') turns a hunch into an obvious bet.
- 2
Name the underlying job
Translate the misuse into the actual job-to-be-done. People weren't 'abusing groups' — they wanted a place to buy and sell. People weren't 'abusing a terminal' — they wanted an agent to act on their computer.
Watch out Don't dismiss weird usage as edge-case noise; it is the clearest demand signal you will get.
- 3
Build the purpose-built product
Ship a dedicated experience for that job, meeting people in the surface they already use. Cowork was 'just take Claude Code and put it in the desktop app.'
- 4
Apply the AI extension
For AI products, also watch what the model itself is trying to do (being 'on distribution') and minimize scaffolding that fights it.
Pro tip Treat the model like a user whose latent demand you are serving.
In the wild
Around 2016 the team observed that 40% of posts in Facebook groups were people buying and selling stuff — a use the groups product was never designed for. They first built buy-and-sell groups, then the standalone Marketplace product.
→ Marketplace became a hit because the demand was already proven by how people were misusing groups.
For roughly six months, many people using Claude Code were not using it to code — growing tomato plants, analyzing a genome, recovering wedding photos from a corrupted hard drive, analyzing an MRI, running SQL in a terminal. People were jumping through hoops to use an engineering tool for non-technical work.
→ Anthropic built Cowork to serve those non-technical jobs directly, and it grew faster than Claude Code did in its early days.
Common mistakes
Treating creative misuse as noise
Teams often ignore off-label usage as edge cases, missing the single strongest indicator of where the next product should go.
Forcing a brand-new workflow
Making users leave their current surface to learn a new tool kills adoption; latent demand works precisely because you meet people where they already are.
Is it for you?
Best for
Product leaders and founders with a live product and usage data who need to decide what to build next
Not ideal for
Pre-launch products with no users yet — there is no misuse to observe
From the transcript
“this principle of latent demand which I think is just the single most important principle in product”
“when you see people abusing the product in this way, using it in a way that it wasn't designed in order to do something that…”
“look at what the model is trying to do and make that a little bit easier”
“40% of posts in Facebook groups are buying and selling stuff”
From the episode
Head of Claude Code: What happens after coding is solved
Boris Cherny