Coding since '88. Now building AI agent teams, omadia.ai & filmflow.pro. Founder @byte5ai . Freediver. Here for what AI does to software, UI & UX.
Joined May 2009
- Tweets 1,962
- Following 0
- Followers 457
- Likes 0
179 Photos and videos
Pinned Tweet
Mar 23
Apps are so dead. Interfaces are becoming side effects.
4
11
1,766
Jun 15
Best open-source reasoning model right now?
24%
Qwen 3.6
45%
DeepSeek V4
16%
GLM-5
15%
Kimi K2.6
165 votes • 2 days
5
4
44
155,263
Jun 13
Does the brain prove scaling isn't the only answer?
LLMs have 10,000x more working context than your brain.
Your brain has 130x more connections than any frontier LLM.
Same units. Wild gaps.
2
1
35
167,998
Jun 13
Well, Fable 5 lasted two days for non-US builders. Now what?
34%
Opus. Loyal... for now.
32%
Off to OpenAI or Gemini.
18%
Going East, I guess,
16%
Self-hosted. Finally.
62 votes • Final results
2
7
85
211,589
Jun 10
Ha, turns out the fix for agents that can't count is embarrassingly old: a calculator. But this is a full spreadsheet, in memory, never rendered. The model thinks, the sheet counts. Agents stopped inventing numbers the day we implemented this: omadia.ai
LLMs guess at math.
So our AI agent platform omadia gives agents a headless Excel in memory. Data goes in, the spreadsheet runs the formulas, the model sticks to reasoning, the agent reads the results back. File's there if you ask.
Take the guesswork out: omadia.ai
4
4
79
244,320
Jun 8
Still running workflows by hand? Repetitive tasks eating up your day? Many of them can run on an agent. With omadia.ai you don't need a dev for this. The agent builder does all the heavy-lifting. Oh, and it's free and open-source.
Your team answers the same questions 20 times a week. 🙄
Build the agent that answers and acts. Right where they work. 💡
✅ No code. No vendor. No black box.
✅ Self-hosted. Your LLM keys. Your data. GDPR-ready.
✅ Free. Open-source. Made in EU.
➡️ omadia.ai
3
4
11
94,222
Jun 8
The builder timeline is getting recursive.
We ask what people build to build tools that find more people asking what people build.
Anyway. What did you build? Or tried. Or gave up on. Drop a link.
2
9
57,154
Christian Wendler retweeted
Jun 7
Agent is mid-run. You realize it's heading the wrong way.
Until now: watch it finish, then correct.
With Omadia v0.7: type a message while it's running. Injected at the next iteration boundary. No stop, no restart.
omadia.ai
1
4
157
Jun 7
Your company's AI agents. What's actually true?
27%
Works as promised
27%
Demo gold, prod hell
25%
Deployed, unused
21%
Slideware
307 votes • Final results
4
26
85
328,883
Jun 7
Running a social media account takes more tools than it should. Buffer or Hootsuite for scheduling, a separate analytics view, something for writing drafts. Multiple subscriptions, still context-switching.
Claude Cowork handles all of it. Just prompts. Here is the setup for X.
1/8
11
18
234
516,933
Jun 7
6⃣ ALGO LAYER
X's algo predicts 15 engagement signals per post. P(reply), P(repost), P(quote) and more.
This open-source skill scores your draft against those signals. Shows which ones you are hitting and what to change for better reach.
github.com/iret77/x-post-opt…
7/8
1
161
Jun 7
✅ That's a starting point. The X connector needs a one-time API setup. Everything after that is just prompts inside Claude Cowork.
✅ No subscription tools besides Claude and X, no fixed feature list. Build exactly the social media workflow you actually need, not the one someone else decided to ship.
💟 Like and repost if this was useful.
8/8
1
1
115
Jun 6
The phrase "Hey builders and founders" is slowly becoming X's version of "to whom it may concern". And yes, I am probably in the target audience.
5
22
166
194,426
Jun 5
Can we agree that the most dangerous AI bugs are the quiet ones?
An agent shares something from the secret meeting. A tool returns one field too much. The model casually leaks it in the answer. Congrats, your agent is now a compliance incident.
Marcel wrote about the parts of omadia.ai that had to work securely before this could be used in real teams. Highly recommended read.
9
10
163
648,247
Jun 2
Imagine discussing something with your teammates while an agent is in the session too.
The agent takes notes. But those memories are only retrievable by the people who were there.
This is the point of omadia's multi-tier graph memory system: shared agents can work in teams without turning every memory into shared knowledge.
It keeps private context private.
omadia.ai is the open-source AI harness for grown-ups.
2️⃣ Agent memory breaks when every memory is treated as public.
omadia.ai uses access-aware graph memory so agents only retrieve memories the current user is allowed to see.
That makes shared agents work in real teams, not just private sessions.
1
3
34
133,035