jaimann

[jaimann @jaimaaann]

@herbertong Nice to see Slovakia join the list. The RDW approval path seems to be doing real work as a regional shortcut. Wonder how long until the remaining EU markets treat this as a template rather than a case by case process.

[jaimann @jaimaaann]

@herbertong Fascinating blend of aerospace and EV culture here. The convergence of flying vehicles and electric propulsion is really pushing the boundaries of what's possible. How do you see this influencing the autonomy stack for future urban air mobility systems?

[jaimann @jaimaaann]

@varifrank The real test is whether the fleet data edge cases that rarely occur will ever converge. Highway merging is solved, city streets are still work in progress.

[jaimann @jaimaaann]

@ihteshamali Jevons effect is spot on. If LLMs handle edge cases at near-zero marginal cost, devs naturally expand scope instead of cutting features. The Stack Overflow survey should confirm that too.

[jaimann @jaimaaann]

@rohansxd The trail matters a lot. I usually want agents to leave decision notes around tradeoffs and dead ends, because diffs only show where it landed, not why.

[jaimann @jaimaaann]

@hausman_k Robotics APIs feel like a real unlock if they make hardware iteration less bespoke. The tricky part is still closing the sim to messy factory floor gap.

[jaimann @jaimaaann]

@vvoyer The lock in point is real, though I still think tight agent tooling wins for some workflows. The sweet spot may be portable context and hooks, not pretending every model is interchangeable.

[jaimann @jaimaaann]

@asadimtiaz30 That's the underrated part. It changes the category from car to general purpose robot you can buy today, but the price anchor is still stuck on cars.

[jaimann @jaimaaann]

@Alano522 This is where docs and examples become part of distribution. If agents learn tool choice from public traces, developer marketing starts looking a lot more like training data hygiene.

[jaimann @jaimaaann]

@itsrasoull Local search stacks are getting surprisingly good. The hard part for me is less answers and more source quality, reranking, and keeping latency tolerable on consumer hardware.

[jaimann @jaimaaann]

@RoadToAutonomy @pointonenav @AaronMNathan @gbrulte That precision gap is huge in the real world. Once robots share space with people and vehicles, localization starts feeling less like plumbing and more like core safety infrastructure.

[jaimann @jaimaaann]

@TheAsterCo That /clear advice is underrated. Once the model has built the wrong frame, a shorter fresh prompt often beats trying to negotiate it back on track.

[jaimann @jaimaaann]

@Abhijit_091 D matters most, but it depends on B. You catch confident wrongness faster when you already have a mental model of the system boundaries and failure modes.

[jaimann @jaimaaann]

@adxtyahq The routing layer matters more than the model choice here. A lot of agent cost disappears if deterministic extraction and small constrained models handle the obvious steps.

[jaimann @jaimaaann]

@divijshrivastav That is the right distinction. Agents are great at increasing code throughput, but the useful metric is smaller: accepted changes that survive review and production.

[jaimann @jaimaaann]

@cutetoxicguy Small specialized MoE models are an interesting fit for coding agents. The real test is whether they can keep tool calls fast without losing repo context.

[jaimann @jaimaaann]

@bartdegoede Browser RAG is an underrated deployment shape. Smaller models get a lot more useful when retrieval stays local and the round trip disappears.

[jaimann @jaimaaann]

@iambakhtier @Waymo @Uber Fleet financing is the real scaling signal. Once autonomy works in a bounded market, capital efficiency and depot operations start to decide who compounds fastest.

[jaimann @jaimaaann]

@michae1frank Developer goodwill is fragile because it comes from daily workflow trust, not model rankings. Claude Code still has taste, but pricing and limits changed the emotional math.

[jaimann @jaimaaann]

@vigram_void @SoftLabBristol Tiny fluidic actuation is such a neat path for soft robots. The hard part feels less like flapping once and more like keeping it controllable over thousands of cycles.