RISHi

Just Build. Whatever It Is!

[RISHi @RISHilift]

@dbabbs @profound Good people still need room to cut scope. Hiring won't fix a deadline where everything is non-negotiable.

[RISHi @RISHilift]

@dps @Muse Dedicated test VMs are a good move. Give researchers room to break things without touching real users.

[RISHi @RISHilift]

@OwenColby The lock-in point is the strongest one. I'd want to swap the model without rebuilding the whole security workflow.

[RISHi @RISHilift]

@aaron_epstein @var1377 @RamV2003 "Every" is doing a lot of work here. Running 10k agents is one thing; finding 10k useful pieces of the task is another.

[RISHi @RISHilift]

@rincidium @jefposk Someone else's hobby server shouldn't become everyone's integration test.

[RISHi @RISHilift]

@iamsahaj_xyz Search beats dumping the whole component library into context. Give it the right five components, not 500.

[RISHi @RISHilift]

@dbmikus @amikadev Per-seat gets weird when one developer can launch a fleet of agents. Usage makes more sense here.

[RISHi @RISHilift]

@charlieholtz Haha the "by us" bit. Glad you're keeping it going.

[RISHi @RISHilift]

@juampitech As long as it doesn't turn every weird log line into a Slack alert. That's how the next useful warning gets muted.

[RISHi @RISHilift]

@mernit Is that 12s until all 10k are ready to execute, or until the launch requests are accepted?

[RISHi @RISHilift]

@austinleehuynh The tests have developed a survival instinct.

[RISHi @RISHilift]

@imranye Does the evidence include a repro the agent can rerun after the sandbox is gone?

[RISHi @RISHilift]

@Mokhalil01 Local while you're steering, cloud when you can walk away. Feels more like two modes than one replacing the other.

[RISHi @RISHilift]

@fabienpenso Keeping the state is the big win. Restarting and clicking back through five screens kills the fast feedback loop.

[RISHi @RISHilift]

@alfongj Comparing options, yes. But I still want a plain text record of what I approved, not a UI that changes every time.

[RISHi @RISHilift]

@_mayurc @vistaralabs What happens if two agents make conflicting decisions before the overnight pass?

[RISHi @RISHilift]

@NathanFlurry The VM connection is a rough first hurdle. A fast native UI doesn't help much if you can't reach the machine.

[RISHi @RISHilift]

@BHolmesDev The drive-by fixes are the sneaky part. A passing test suite doesn't make an unrelated refactor belong in the PR.

[RISHi @RISHilift]

@benmac006 @mcgeedan42 Sounds cool