Ümit (@umitunal_) profile picture
Joined January 2011
Photo by Ümit (@umitunal_)Photo by Ümit (@umitunal_)Photo by Ümit (@umitunal_)Photo by Ümit (@umitunal_)Photo by Ümit (@umitunal_)Photo by Ümit (@umitunal_)
I’ve been thinking about how AI agents understand and use data. This article offers an interesting perspective on why ontology is becoming important again. Worth a read! aiexpjourney.substack.com/p/…
3
1
15
1,285
Ümit retweeted
Mental models that runs almost 93% of engineering
45
1,457
27
9,006
323,857
Ümit retweeted
herdr-core was built for this. tell your main agent to read hc - - skill and orchestrate all other agents and terminals for you!
I've changed how I work over the last few weeks. I now have one lead agent for each area of focus, running all the time, and a chief of staff agent that oversees them. Each lead manages its own sub-agents that write code, test it, chase reviews, and coordinate with team members. Everything I need to decide or review comes into one prioritized queue, a simple local web app. Each product review includes screenshots and a video demo of the feature. I clear the queue in batches, mostly by voice and keyboard shortcuts. The leads also bring back ideas I hadn't considered. Working from one queue, instead of switching between dozens of chats, has cut a lot of mental overhead. The system keeps learning my preferences and distilling my judgment into a "Brian brain," so its recommendations and first drafts get better before they reach me. The goal is fewer corrections per change. Every change still goes through the same human review and security checks. It is eliminating tasks, not human judgment. What an exciting time to push the frontier of productivity, especially as we build the economy for agents (agentic finance). We should all keep pushing in this direction.
40
27
8
1,200
82,413
Ümit retweeted
Life hack if your flight doesn't have Starlink: ask your agent to bypass the 443 split-TCP proxy via a $0.03/h BBR multiplexer and get ~4× faster, reset-free internet. Open-sourcing it from the clouds: github.com/Th0rgal/satmux
70
132
24
3,632
268,258
I just published Stop paying for BI seats: build a self-updating dashboard in Google Sheets. @thunderduckio medium.com/@yadiguzel/stop-p…
2
2
145
Ümit retweeted
Andrej Karpathy: “A 1B model can be much more effective than most current LLMs. 90% of current models waste a ton of capacity and money on remembering stuff that doesn’t matter.” here is his 5 core thesis about Small LLM’s: 1. the intelligence was never the knowledge. Strip the memorized internet, keep the reasoning. What’s left is the cognitive core, and it’s small. 2. the core might fit in ~1B params. Maybe less. “Plenty of room at the bottom.” We’ve been paying for a library when we needed a brain. 3. size stopped being the bottleneck. What makes an agent work is the tools it can call and how tight the feedback loop is, not parameter count. 4. intelligence lives in the system, not the weights. A raw LLM is a CPU with no OS. Small core + good harness > giant model alone. 5. you can own it. A model on your hardware is one no one can rate-limit, price-hike, or deprecate on you. in 12 months the gap between teams shipping on-device and teams paying per-token frontier bills won’t be capability. TwIL an 8.8B model at 90%+ of frontier isn’t a benchmark flex. It’s the thesis Karpathy’s been repeating all year, made real. study more on this below ↓
Modest intelligence is arriving faster than most people realize. Today, we're open-sourcing TwIL-8B. Just 8.8 billion parameters, yet approaching frontier-model scores on several difficult benchmarks: • 93% of frontier performance on coding • 92% on instruction following • 90% on contest mathematics All from a model small enough to run on your own hardware. But the benchmarks aren't the most important part of this announcement. For years, the AI industry has focused on building the most capable model, regardless of the cost required to run it. I believe that equation is changing. The next era of AI will be defined by the cost of completing work. Not simply the intelligence of the model, but how efficiently that intelligence can accomplish the task. If a model with 8 billion parameters can reliably complete the same work as one with hundreds of billions, why pay for the larger model? As smaller models become more capable, intelligence becomes something we can distribute, specialize, and run across the hardware we already own. That's the future we're building at @thewebAI TwIL-8B is available today. Open weights. And this is just the first of several announcements we're making today. HF Link: huggingface.co/webAI-Officia…
203
658
135
7,001
1,288,866
So much value in the skill of: - Grab a book - Turn off distractions - Focus, read, and learn
33
68
9
1,302
45,196
herdr.dev became a real Mac app! I can't stand the terminal, so I've been waiting for this! 👌
34
9
3
326
45,009
S3-based database architectures don’t read from S3 all the time, though. Instead, they amortize the cost by caching in memory and on local disk (which can be NVMe). As a result, read performance is more or less the same as in local-disk architectures, though the tail latency is heavier; NVMe’s tail is much flatter. Writes are slower, but not by as much as people often think. You almost never want a write acknowledged by a single local disk, so a local-disk database pays a network round trip for replication anyway. That makes S3-based architectures competitive, especially with S3 Express.
TL;DR your database should be on local NVMe
25
22
6
385
44,243
Ümit retweeted
The 10-year database architecture cycle: 2015: SQL is obsolete. Schemas are too rigid. Put everything in NoSQL JSON documents. 2018: Wait, we need to join users and orders. Let's write a 400-line nested loop in JavaScript to join them in RAM. 2021: Wait, if the server crashes halfway through the loop, we lose money. We need ACID transactions. 2026: Quietly puts everything back in PostgreSQL and never speaks of it again.
141
430
44
5,227
267,113
Ümit retweeted
*ehm ehm* BIG ANNOUNCEMENTS DAY 📣 two new things cooking in herdr labs 🐏🧪 a brand new herdr-core, and a desktop app built on it. open source. first, the app: software that fits you. every part of it is a file your agent can change. you just ask.
297
413
278
4,737
503,620
This is seriously impressive. Amazing work! 🔥
I built bbtree, the foundation of databases and filesystems, and the fastest concurrent btree. Code: github.com/XiangpengHao/bbtr… Blog post & demo: bbtree.xiangpeng.systems/
5
213
Ontology-Driven Observability: Building an End-to-End Knowledge Graph at Netflix Scale The slow part of incident response is reconstructing context, so write system's entities & reasoning down as a machine-readable model once & let queries do the joining netflixtechblog.com/ontology…
4
26
166
11,510
I am a bit of a SlateDB fan. A few friends recently asked me what it is, so I made this short intro. I hope it helps more people learn about SlateDB. Thank you, @criccomini , for building something so cool!
3
9
1
103
7,006
There's an obvious way to build a database router. We didn't do that. Nothing's faster than doing nothing. See how we built Neki with the laziest Postgres router possible. planetscale.com/blog/designi…
9
3
2
82
7,844
10 Years of the SRE book: Announcing the second edition of "Site Reliability Engineering” cloud.google.com/blog/produc… < I'm excited about this and glad to see this foundational book updated with evolved SRE practices and AI factors.
2
95
1
489
20,116
Ümit retweeted
We've always called it software engineering, but we've actually been builder...gineers. In civil engineering, aerospace, or architecture, the people who do the engineering and the people who do the building are obviously different. But in software, we collapsed them into the same role. With agents, we're finally separating the jobs. We're going to specify and encode more upfront, add more guardrails, and design architectures that bias agents toward success. In many ways, we're finally starting to look like actual engineers.
18
25
6
266
10,115
It's a pity JVMCI was removed from OpenJDK now that Graal CE JIT gets faster than C2. Graal CE 25.4.4 Native Image performance improved significantly. The last missing feature is probably PGO. Shenandoah for NI is coming along nicely too #graalvm @graalvm github.com/simonis/Renaissan…
2
20
2
52
2,788