Introducing Ask Tile.
Most AI tools require you to explain what you're working on before you can get anything done. With Ask Tile, that step doesn't exist.
Click on any element in your app: a button, a chart, a form, and an AI assistant opens with full context already loaded. Just ask for what you need. That's it.
It's a small shift in workflow, but once you've used it, going back feels painful.
Hot take: bugs don't slow you down.
Slow feedback loops do.
The real cost of a bug is the distance between when it’s introduced and when it’s observed.
Preview app changes in real environments before production. Surface issues early. Fix them fast.
Code generation is commodity. Top coding models have similar performance on coding benchmarks.
Agentic coding is the skill devs need to master. It’s going to radically transform engg orgs.
Building a mobile app isn’t hard.
Shipping a reliable, scalable one is.
We close the gap between both.
We’re building https://t.co/Fst5cVNmdO for teams shipping real mobile apps.
Not demos. Not experiments.
Apps that need reliability, scalability, and a clear path from idea → production.
@naval Nope. UI is literally user interface. Most of it is visual. That maps to the way users are wired. Human brain is optimised for visual information processing. Text or audio doesn’t come close.
@karthikkalyan90@arafatkatze@cline We use a combination of sliding window and summarisation. Now adding focus chain. Offloading context to file system or alternate persistent storage sounds promising.
@GergelyOrosz This happens more often than people will admit. Making engineering leadership own outcomes and working closer to sales/customers works out well for small teams.
omg... no-code was supposed to help you launch faster
but it never got you to the App Store.
what if you could build a real app this week
and start making money from day 1?
mini story + what actually worked for me 🧵
The bitter lesson is coming for app gen and the best teams are already thinking about which investments are durable vs fleeting. More:
- Retention today is proportional to success rate - did the user get what they expected in a reasonable number of tokens (aesthetics / features / bug fixes / deployment)?
- Many smart teams have engineered around the constraints of models - context mgmt, parallelism, system prompts, design systems etc and are seeing real alpha as a result
- New models will be a tidal wave that overwhelms much of this work - teams will need to be ruthless and avoid sunk cost thinking + building new sources of alpha as needed
- The best teams are already thinking about durability - integrations, re-usable components, and routing inference / payments feels safe; the smartest thing I heard this week was “we’re leaning into the bitter lesson”
- The only thing I’m sure about at this point is that the world has 1% of the software it will have in 5 years
If you’re building in the space I want to hear from you - anish at a16z
@illscience@tile_dev building fleet of specialised agents that build native apps and ship on our deterministic, managed infra. Future is agentic composition.
We just updated our core framework and agent orchestration system that reduces token consumption by over 30% for all cases. Any errors will self correct in most cases. Effectively, you will spend less than half the tokens for the same outcome.
The biggest ask from builders on @tile_dev is to improve token consumption. The team is working on an update that will significantly reduce token consumption while working with planner agent.
@ItzSuds On a low base this works. B2B is expensive and scale of B2C even at today’s costs works if you can set the funnel to mid range ACV qualified trials/demos