the fact that nobody with decisionmaking power saw this coming a mile away and vetoed the 6.2 astra ad is my primary critique of the astra launch
the model itself is solid minor version improvement guys, efficiency is one of the three most impt RL objectives rn and they are world leading here!!! i am daily driving astra and it is the best workflow automation model i’ve used, no longer getting stuck in horrible loops and efficiently learning thru compactions and skill writing. it’s running a lot of our content publishing pipeline for @latentspacepod and @aiDotEngineer and we wouldnt have nearly the revenue efficiency (>1m/employee) that we have as a bootstrapped business without it
technical details:
> inter-session context is surfaced by default. Teams now create dedicated bots for different projects; while these different devins can talk to each other when necessary, they each have their own individual memory bank and goals. Each bot now has its own, long lived identity. The transition from "Devin is a swarm of ephemeral agents" to "Devin is a collection of bots, each with their own persistent identity and memory" is changing how people use and prompt Devin.
https://t.co/maw20vdPs8
devin bots are multi-agent teams done right:
- always on, proactive (but tuned for ROI)
- own outcomes not tasks (but can collab w others)
- manages teams of subagents (and reviews itself)
- available everywhere (call text slack whatever hello @interaction)
- consistent reporting thru interactive artifacts aka devinapps not long chats (HTML is all you need™ @trq212)
this is the preview that @walden_yan gave at AIE. i've been trialing them for a week and i've been able to hand over my AEO, QA, cross-platform app porting, personal admin automation, and video editing work (also simultaneously moving from local agent worktree hell to cloud because ofc this is all cloud based thank god)
hyped concepts this replaces:
- /goal
- "ultracode" / "ultra mode"
- "loop engineering"
- "agent to agent"
- "agent swarms"
- AI employees (see below)
my main aha moment is that you don't want "AI Employees", we do not have agi yet and you will just fail if you overpromise/overexpect. what you want is the "just" the next level of abstraction above single tasks, and the overwhelming consensus I've had talking to everyone at the frontier of agents (yes incl cog competitors) is agent managers that own outcomes.
incidentally, as an employer of humans it's also how i distinguish between someone who "has agency" (owns a problem area, finds issues and researches/brings solutions not problems) and who doesn't (only does precisely what i asked, not proactive, doesn't come up with suggestions i didn't think about). guess which one i value more lol.
if youre a startup, reminder that Devin for Startups offers 65k in credits. apply via clanker