Carl Lapierre (@dev_clap) header
Carl Lapierre (@dev_clap) profile picture

Staff Developer at Osedea Agentic Systems & Engineering

Montréal, Québec
Joined December 2016
funny how much simpler agents have become. we went from recreating entire hierarchies to letting a capable model handle more of the work in a simple harness. @thsottiaux talks about this with @GergelyOrosz some of the harness “crutches” become unnecessary as models improve. worth revisiting what your architecture is still compensating for. stay close to MCP, ACP, skills and plugin conventions so the pieces are easier to integrate and move between systems. reassess the orchestration every few months, etc..
22
How cybersecurity looks pre-AI
19
Replying to @zachlloydtweets
Agreed, I've been working on a software factory methodology and I always come back to this critical component of earned autonomy. It's the main reason why people are skeptical about SFs in the first place as it's assumed a production line will be able to take on any type of task.
253
Data entry is about to change so drastically
17
Replying to @ThePrimeagen
This is my embodied autonomous agent with computer use meant to stream on Twitch. Powered by Codex & Pi
7
654
Last night, at @AITinkerers Montreal, I demoed "Sleazy Ron", a dual agent system powered by Pi Agent Harnesses and Codex to live stream on Twitch. Pretty happy with the results so far. gpt-5.4 is genuinely funny.
6
3
11
3,305
Why is Codex so sassy with me all of a sudden. It got annoyed that I deleted its generated assets before commit. 😅
67
Lately, @steipete and Codex have made me want to build fun UIs again. This is going to be an @openclaw agent sandbox for multi agent collaboration. I don't know how its going to work but it should be fun! The sandbox itself is quite polished now.
203
Pretty crazy where all of this is going, this was just one of my small test using Kling AI motion control. The world of entertainment is getting disrupted in real time.
109
I've been trying to get LLMs to generate pixel art token by token with many different models, @Google s gemini 3-pro is the first to successfully one shot art at 8x8 & 16x16 👏 I guess this is my new benchmark for language models.
1
147
Replying to @cursor_ai
@cursor_ai and I just vibing today
1
1
26
This is sick! The unofficial #vibejam party just played a round of FoodVibers by all jumping in from the Vibe Metaverse! It's so fun to play with so many players at once! Just missing @levelsio @martyamark @StefanBuilds @alightinastorm @swe_acc @m_davies94 ++
2
5
1
31
1,553
I've been building this GraphRAG visualizer on the side recently. I think it looks really cool explore. This specific dataset was generated by @grok about movies, you can sort of see the how synthetic it is.
2
6
476