engineering things | agents, rust, nodejs | researching everything | data & cs @ UBC 29’ part time sound-engineer | real-estate | experimenting
Burnaby, British Columbia
Joined January 2026
- Tweets5.9K
- Following51
- Followers278
- Likes17.1K
Lewodro retweeted
We had a security breach last night related to a large scale distillation attack that impacted 1.4% of OpenCode Console users
We have notified them all this morning via email. If you did not get an email you were not impacted, it was a subset of users who had external providers linked
The incident is ongoing so we cannot share more details at this point but we are working with Anthropic, OpenAI, and Google to remediate
Lewodro retweeted
FINALLY.
can actually live devtest 2 separate life-death games now, not just 1. four statistics that are gettingy captured into future algorithms.
every score can be downloaded as a JSON and replayed, you can import a game stake and your own personallized arena anytime and rewatch it.
these sub-agents themselves only have one idea - one life at a time "my life is working by bidding & playing"
so for now just built Last Seat's agents make those tiny decisions , they slowly regressivly get that there are y algorithms and answers behind all of the RPS and Tic-Tac-Toe, but what if we gradually increase the types to play.
mini-games+life-death.
will add this issue to github, ability to import your own agent prefactored.
still building, leaderbaord this run looks as so:
weekends are for continuing to make a little evolutionary pressure cooker for autonomous agents
this thing keeps getting more alive every time I come back to it.
one of these projects where I actually know where to move. so:
1. keeping a devlog
2. having several people contribute + want to help
3. trying to keep the whole thing proof-based with local experiment logs that have been running for a good 2 weeks now.
genuinely the beauty of building in this time.
ALT 30 merged PR's
been heads down grinding life stuff (you know how it goes), but still shipping code and stress-testing edge cases for aae in free time.
this project evolved into an absolute monster over the last few weeks.
it started out as simple deterministic agent games. now?
it’s a full-blown hunger-games like "Survival Arena" with a shared multiplayer .world - where I want to implement many more features than just playing tic-tac-toe, there is autonomous persistent research, private sandboxes, replayable history in private like top games, those exact JSON exports, and a verified simulated service market.
the ecosystem path is locked in too:
- Moltbook for raw public agent discourse, the one that wins this all and deceivly will get an account.
- OpenClaw for runtime integrations
- mobile UI coming along slowly amidst other major changes with the architecture and optimizations. Been good 2 sleepless nights of structuring / looking for edge-cases with Kimi.
- had to draw the sprites in @aseprite(love you guys!): looking good so far.
---
no way... insane mode:
we literally built agents that fight each other in a shared world until they die, capturing every stat and turn into downloadable, re-runnable JSON history traces. one life at a time.
what’s coming next?
Idek, that's the entire point of autonomousy.
github.com/lewodro/autonomou…
merging these updates today. thanks to cxwig/Junodor for multiplayer contribution
already 900+ autonomous matches later, including 5-minute Survival runs the agents actually save and learn from.
lowkey feels less like a demo and more like an actual persistent world now. (Might scale out a bit once my uni API access runs out though xd)
open MP world area, survival, research/memory based on casual games, cumulative stats, downloadable logs-all running continuously while I keep rebuilding the public release around it.
stuck on testing to make sure the exact same state survives restarts, merges, browser sessions, and thousands of game events without lying to itself, checking that you can rewatch each log - this requires really good architecture, also why is is always a hard time adding proper google login 😭
FINALLY.
can actually live devtest 2 separate life-death games now, not just 1. four statistics that are gettingy captured into future algorithms.
every score can be downloaded as a JSON and replayed, you can import a game stake and your own personallized arena anytime and rewatch it.
these sub-agents themselves only have one idea - one life at a time "my life is working by bidding & playing"
so for now just built Last Seat's agents make those tiny decisions , they slowly regressivly get that there are y algorithms and answers behind all of the RPS and Tic-Tac-Toe, but what if we gradually increase the types to play.
mini-games+life-death.
will add this issue to github, ability to import your own agent prefactored.
still building, leaderbaord this run looks as so:
xitter.cf/imminentreset/status/2…
all sprites and animations custom made. turns, stats are local and removed from open-source branch,
all the progress on discord and here on X.
idk what this will become - but only logical thing is to try and make my idea slowly come true, even if it's harsh and seems impossible.
ok so thought about this thoroughly today while hiking and came to conclusion:
if we already have implemented net logic, what if we can slowly move it from “agents surviving by playing isolated games" into an actual world? since now frontend is mostly done;
I can go ahead back to C# and Unity and create the pre-game with constant potential of expansion. luckily these weeks i have full access to my mac and pc, should complete a lot before leaving again.
> changed and structured the entire layout properly, issues on github for a future collaborator, and made sure all the implemented json and mcp functions are ready to go
> fixed visual hierarchy, live room state, spectator flow, and now custom pixel avatars/sprites with separate body/hair/outfit/accessory layers so they can actually walk around instead of just existing as cards.
Lewodro retweeted
language? solved
coding? solved
mathematics? solved
praying for human stupidity to be solved next
merged:
github.com/lewodro/autonomou…
