Live in the Future, But Not Too Far
Hold the far-future vision, but land with users where they already work today.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 85%
Teams building at the frontier can build products so far ahead that users can't adopt them. The fix is to keep the ambitious end-state vision while shipping an intuitive on-ramp that meets users in their current workflow, then gradually pull them toward the future. It also warns that internal dogfooding signal can be misleadingly ahead of the general market.
Origin
Alexander Embiricos drew this from OpenAI's experience launching Codex Cloud (an async cloud agent) that internal power users loved but the broader market found too hard to adopt.
Core principles
- 01Frontier teams naturally over-build for the far-future power-user journey
- 02Your dogfooding signal can diverge sharply from the general market's readiness
- 03Adoption requires an intuitive, trivially-valuable first touch before the visionary workflow
- 04The future vision is still correct — sequencing, not direction, is the error
How to run it
- 1
Keep the far-future vision explicit
Hold a clear picture of the end state you're building toward (e.g. a proactive agent teammate you delegate to), and don't abandon it.
Pro tip Reason about roughly year-plus horizons as fuzzy aim, and weeks/low-months as concrete — the awkward middle near one year is hardest to plan.
- 2
Check whether your dogfood signal matches the market
Notice where your internal users' habits (e.g. comfort with async, parallel, reasoning-style prompting) make them unrepresentative of outside users.
Pro tip At OpenAI everyone is used to firing off parallel reasoning tasks and returning later — most of the market isn't there yet.
Watch out Dogfooding at a frontier company can make a too-early product look validated.
- 3
Land with an intuitive, trivial-to-value on-ramp
Ship the version that plugs into how people already work (IDE extension, CLI) so they get value immediately without learning a new paradigm.
Pro tip The key unlock is making the first experience intuitive and trivial to get value from, not maximally capable.
- 4
Convert the on-ramp into gradual movement toward the vision
Design the everyday product so that using it naturally configures it for the more autonomous future workflow you actually want people to reach.
In the wild
The cloud-first async agent was a massive internal accelerator but hard for the market to adopt because of environment setup and unfamiliar prompting. OpenAI kept the delegation vision but led publicly with the local IDE/CLI experience.
→ Codex grew 20x since August 2025 and became the most-served coding model in the API after leading with the intuitive on-ramp.
Common mistakes
Shipping the visionary workflow as the front door
When the first experience demands new paradigms (async delegation, environment config), users can't cross the adoption gap even if the vision is right, and the product stalls.
Trusting internal dogfood enthusiasm as market validation
Frontier-company employees are unrepresentative power users; their love of a product can mask that the general market isn't ready, leading you to over-invest in the wrong sequencing.
Is it for you?
Best for
Product leaders at frontier or fast-moving companies whose internal users are far ahead of their target market
Not ideal for
Teams whose users are already sophisticated adopters, or incremental products with no big paradigm gap to bridge
From the transcript
“the signal we got from dogfooding is a little bit different from the signal you get from like the general market”
“the key unlock is actually first you need to land with users in a way that's like much more intuitive and like trivial to get…”
“It's like live in the future, but maybe not too far in the future”
From the episode
Why humans are AI’s biggest bottleneck (and what’s coming in 2026)
Alexander Embiricos (OpenAI Codex Product Lead)