Building Physical-AI, operational and situational awareness platform! Glass is half full kinda vibe. ex X, Meta, Unity, Dell. Founding member of 2X startup.
San Francisco 🔁 Vancouver
Joined January 2015
- Tweets182
- Following610
- Followers163
- Likes1.9K
Starting a new venture with a friend in fintech. Looking for founding engineer and deisgner. Prefer:
- based in vancouver or around(will consider remote canada if absolutely exceptional)
- high agency
- ai-pilled
- system architecture
Work-life balance isn’t a thing for founders. But oh boy, a loving family has a way of filling your tank like nothing else. A few moments of pure joy with your kid, and you find yourself walking back to work with renewed energy and an even stronger sense of purpose.
I read @lilianweng's
xitter.cf/lilianweng/status/2070… and this feels like the operational version of scaling laws. In model training, bigger isn’t always better, the balance between model size, tokens, data quality, and compute matters. In AI adoption, it’s the same principle: don’t suppress usage, reduce waste through better defaults, routing, caching, and visibility.
How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching.
Better Defaults (not Usage Caps) – Engineers can choose any model they want, but defaults matter. We’re experimenting with defaulting to open weight models like GLM 5.2 and Kimi 2.7 through our LLM gateway, while still encouraging engineers to choose the right model for the task. 91% of our employees were never hitting their usage caps, so instead of lowering caps and driving up alerts, we're moving to cheaper defaults. Note that code reviews use a diversity of models, so they can check each other's work.
Better Routing – In our custom harnesses, we preprocess prompts and route to the best model for the job, considering cache hits and model pricing. For instance, you may want a frontier model for planning, but not for execution where they can be overkill. Ultimately, humans shouldn't be choosing models - AI can automate this task.
Better Caching – Cache misses are the easiest way to drive your cost up. All of our requests are cache aware, so we’re reusing a warm cache wherever possible. For example, our cache hit rate went from 5% → 60% in LibreChat once properly implemented.
Keep Context Lean – Start fresh sessions when switching tasks. Scope file context narrowly. Disconnect unused tools. Don't just compact. The goal isn't fewer tokens used, it's fewer tokens wasted.
Better Visibility – Our engineers can use as many tokens as they want, from whatever model they want, but we’ve made usage visible – and the more you spend on AI, the more impact we expect.
The goal isn't to suppress usage. It's to build the infrastructure that makes exponential growth sustainable.
Putting this into practice has cut our AI spend nearly in half, while our token usage continues to grow.
The “founders become unemployable” myth died the moment AI made execution cheap. Now companies need founder energy: taste, grit, speed, obsession, experiment loops.
The highest leverage people today aren’t ticket closers, they’re momentum creators.
Ironically, frontier labs are hiring exactly the people old corporate logic filtered out.
Samreen retweeted
if you’re freaking out about Mythos, remember:
Never make any major life decisions within 30 days of a meditation retreat, psychedelic trip, or first encounter with a frontier AI model.
true AGI for me is when i have an agent who can go through all SAP workflows and interact with those aweful UI to finish all tasks........
The smartest AI won’t win. The most useful one will.
Right now, most builders are just accelerating software, tools for people who are already 10x leveraged in white collar jobs.
But the real unlock?
Making everyone else 10x better.
Teachers. Accountants. Small shop owners, supply chain managers, old regulated industries, helathcare professionals and Community leaders.
That’s where adoption lives.
Intelligence is built.
Adoption is earned.
Have been advising early-stage founders and one thing i truely advise specially to non-technical founders is that your first 2 engineering hires define your trajectory. Hire a pirate, someone who can vibe code, move fast, hack together an app, and get something tangible in front of users and investors.
And a captain, someone who understands systems, builds for reliability, security, and turns that scrappy demo into a production-grade product.
Early speed gets you in the door and create the buz.
But only stability keeps users from churning.
AI builders mostly talk to each other.
So they build tools for themselves.
Regular people get left out.
The products feel confusing, overbuilt, or irrelevant.Then the same builders wonder why most of the world still uses AI like a basic chatbot and starts calling it hype.
if you are a builder then get out talk to everyday consumers and non tech users and while you are at it......touch the grass!
Yes, every ai first company must invest in leveraging proprietary workflow data or your agents will be no smarter than the rest, leading to thousands of identical bots churning through monotonous tasks with zero edge.
Larry Ellison $ORCL stressed that AI models perform at their best when trained on privately owned data—not just the public web data used by GPT, Grok, Llama, and others.
That’s why companies like $HIMS, $DUOL, $LMND are set to win.
Proprietary data is the real competitive edge.
ChatGPT can’t match that
Samreen retweeted
Remember this: You are not responsible for the way your growth journey makes other people feel. Your growth exposes their insecurity. They'll tell you to be realistic. They'll belittle your ambitions. It's ok. It's natural. A cost of entry. Pay it with pride.
Samreen retweeted
The saddest thing about success is realizing how few people actually wanted to see you succeed. They smiled and cheered to your face but secretly hoped you’d fail. Learn to recognize that quiet envy. Cut those people from your life. Small circles create big outcomes.
Oh, exciting drop from Google, the File Search Tool in the Gemini API, a fully managed RAG system that grounds models in custom data via automatic chunking, embeddings with gemini-embedding-001, semantic vector search, and prompt augmentation complete with citations. Free storage and query-time embeddings, just $0.15 per million tokens for initial indexing. No infra hassles, integrated right into generateContent.
Props to @GoogleAIDevs: xitter.cf/googleaidevs/status/19…
I’ve been championing this vision of designing agents that transcend stateless, one-shot task completion by incorporating persistent memory mechanisms, interaction-based learning loops, and iterative workflow refinement. Tools like this validate the shift for firms building agentic systems, elevating generic processes into adaptive, data-grounded pipelines that evolve with user feedback and domain specific contexts.
Reflecting on the landscape, many AI architectures start with foundational elements like basic embeddings and vector retrieval, which handle core queries effectively yet open doors for enhancements in persistence and adaptive learning. Picture the possibilities when those foundations expand to include ongoing memory and refinement, it’s where true innovation accelerates. This release feels like a timely invitation to explore further.
Announcing File Search, a fully managed RAG system that lets you ground Gemini in your own data, built directly into the Gemini API 🔍
It’s simple, integrated and affordable.
blog.google/technology/devel…
Samreen retweeted
Announcing File Search, a fully managed RAG system that lets you ground Gemini in your own data, built directly into the Gemini API 🔍
It’s simple, integrated and affordable.
blog.google/technology/devel…