Pinned Tweet
Here are my 51 learnings from 50 hackathons in 2024!
Endless shoutout to all the organizers, sponsors and hackers for driving the pace of innovation in a region where it’s already sky-high. 🙌 🚢
After reflecting on this journey, I’ve grouped my learnings into these key themes:
✅ 1/10 Hackathons: The Heartbeat of Innovation
1. Definition of hackathon: With ideas you don't have, with people you don't know, with tools you haven't worked with: SHIP USEFUL THINGS FAST!
2. Hackathons are the freestyle dance of innovation, unlike the routine exercises of gyms or corporate life. They keep your engineering and entrepreneurial skills razor-sharp.
3. Hackathons are a great way to challenge yourself, step out of your comfort zone, and embrace creativity.
4. Hackathons unleash collective intelligence that can ripple into positive impacts beyond the event.
5. Hackathons are the perfect environment to collaborate in non-corporate, non-academic settings, enabling individuals to bring out their best.
1/7 In @AnthropicAI's own red-team tests of browser-use attacks, prompt injections succeeded 23.6% of the time with no mitigations. With the new safeguards, that dropped to 11.2%.
Even the best model guardrails stop most prompt injections, not all of them. So we should not give the model the keys.
So I built a security-audit workflow where the agents never receives them.
It starts in Slack, runs in GitHub Actions, and produces validated security reports.
github.com/sakomws/slack-age…
7/8 I ran it against @goupaz.
Result:
✅ No confirmed vulnerabilities
✅ Two authorization leads requiring validation
✅ Deferred areas stated explicitly
✅ No fake severity without execution evidence
github.com/sakomws/goup.vc/a…
Just presented Slack Agent Runtime at @PlatformDay Day 🐴
Harness: Slack
Inference+Search: @YouDotCom Inference Router
Data: MCP servers
Bounded Worker: GitHub Runners
Artifacts: Github
The Agent writes the code. It never holds a GitHub or Slack token.
Give the bot the keyboard, not the keys!
github.com/sakomws/slack-age…
Sako retweeted
Kimi K3.1 Pro is live on Inference Router from day 0.
K3.1 is @Kimi_Moonshot's most capable open-weight model yet, with a 1M context window and a reported 65.7% on Terminal-Bench 4.0.
Run it in production with at least 99.9% uptime SLA, US hosting and ZDR available. No upfront commitment.
For last 2 month, what kept me busy🥶
Today is a launch day, great team work! 🥳
Introducing Inference Router: enterprise-grade inference on demand.
Kimi K3, GLM 5.3, DeepSeek V4.1 Flash, Qwen3.8 Flash, Muse Spark 1.3, Mercury 3 and more, available with an uptime SLA of at least 99.9%. US hosting and web search built in.
you.com/router
















