to feel emotionally connected is like decoding a string of bytes
Boston, MA
Joined December 2016
- Tweets786
- Following644
- Followers419
- Likes335
Claude Code Source Code Leaked and The Matter of Open Source open.substack.com/pub/stille…
Docker and Kubernetes were once the kings of DevOps.
But in 2025, their dominance is fading.
If you're still relying heavily on them, it's time to explore what’s next 🧵
What’s replacing them?
🔹 Local Dev: Podman, Colima, nerdctl
🔹 K8s Runtimes: containerd, CRI-O
Not ditching containers - just a leaner, more secure stack.
Explore the tools:
Podman: github.com/containers/podman
Colima: github.com/abiosoft/colima
NerdCTL: github.com/containerd/nerdct…
containerd: github.com/containerd/contai…
CRI-O: github.com/cri-o/cri-o
We studied dozens of GenAI startups.
Most fall into the same traps.
Here are 5 common mistakes and what the best teams do instead 🧵
4. Over-indexing on early success
MVPs are fast. Scaling is slow.
Going from 80% → 95% takes 5x longer than 0 → 80%.
Plan for:
- Edge cases
- Hallucination handling
- Latency tuning
- Model updates
Tweet 6/
5. Skipping human eval
LLM judges are cheap - but blind.
Layer in daily manual reviews.
Human feedback catches silent failures and keeps quality grounded.
We wrote more on this in:
stillenvc.substack.com/p/llm…
What makes AI truly personal and powerful? AI Data Connectors.
They securely fetch and structure your unique data, empowering AI agents to understand your specific context and preferences without compromising privacy or security.
Between 2018 and 2023, AI paper submissions to top conferences exploded from ~12K to over 32K.
But AI is now facing a reproducibility crisis.
As models grow in size and complexity, verifying published results becomes harder - slowing real progress. 🧵
How much AI-generated imagery looks alike? It's a fascinating blend of the underlying technology and how our own human preferences get amplified👇
The real challenge when AI agents built on different systems try to cooperate?
Information asymmetry.
What if we could use blockchain to eliminate these frictions? 🧵
How AI agents think & solve problems? Starts w/ prompt.
→ One-step: get task, output answer.
→ Multi-step: reason, act, iterate to solution.
Key LLM agent prompting techniques 🧵
How to optimize memory in AI agents for smarts w/o huge token costs?
6 practical techniques (basic to advanced) for effective, efficient, affordable memory in AI agents. 🧵
