Jake Loo
@jakeloo
Followers
2K
Following
485
Media
15
Statuses
252
building @thirdweb. @fdotinc. @UCBerkeley. i’ll never message you first on TG - please stay cautious of imposters.
New York, NY
Joined November 2011
it's consistent with what we found ~6 months ago. switching to code execution tools reduces our agents latency significantly https://t.co/TKGpSK3llt
New on the Anthropic Engineering blog: tips on how to build more efficient agents that handle more tools while using fewer tokens. Code execution with the Model Context Protocol (MCP):
0
0
2
4x FASTER. we now default to eip7702 execution for server wallets instead of erc4337. sending to confirmation: 2 secs (~4x faster than 4337!). as fast as a EOA tx, but with gas sponsorship, atomic batching and session keys support!
5
4
32
we just open sourced engine-core 2 years 7 months of our learnings distilled into 25k lines of production Rust. this is the best thing I've ever built as an engineer. our first open source rust repo at @thirdweb. let me tell you about my baby:
13
9
74
making ai multiplayer multiplies intelligence it forces clear communication, and that will lead to better results this will be the norm for how work gets done
We made Claude, but multiplayer. For the first time (ever) -- you can collaborate with a model + others in the same exact chat. Demo + available now:
1
0
10
This is so exciting - LLMs, but for the physical world V-JEPA 2: real-world video in -> next action out LLMs: text in -> next token out Combining them is gonna be wild: LLMs as brains, V-JEPA as bodies. Robotic agents that plan and act are almost here
Our vision is for AI that uses world models to adapt in new and dynamic environments and efficiently learn new skills. We’re sharing V-JEPA 2, a new world model with state-of-the-art performance in visual understanding and prediction. V-JEPA 2 is a 1.2 billion-parameter model,
0
0
6
As companies scale up, they ship more slowly. Why? Coordination. It requires multiple people to ship. An AI agent works across the stack. From reading code and product specs to producing content. Zero coordination needed No more "finding time on your calendar". It just ships
The wildest AI takes are "it's not good enough yet" or "it can't do X well." This completely ignores the exponential improvement curve we're on. If you don't believe every aspect of digital work will reach human-level quality soon, you're setting yourself up for failure.
0
0
6
Interpretability gives users direct control over AI systems to get their tasks done
We created a canvas that plugs into an image model’s brain. You can use it to generate images in real-time by painting with the latent concepts the model has learned. Try out Paint with Ember for yourself 👇
0
0
6
The development time has gone down so much that it's more impactful to just ship, learn and iterate
2
0
12
Eval-driven development (EDD) is the new test-driven development (TDD)
1
0
2
There's a big shift happening in how we learn online Search-based learning required knowing what to look for and refining queries yourself LLMs make learning more collaborative and exploratory - they help you discover the right questions “good at googling” is now irrelevant
0
0
3
IT'S DEMO DAY 🎉 Come see 30 of the most exciting web3 startups demo their apps, including ◆ Onchain AI Agents ◆ Blockchain games ◆ Consumer crypto apps https://t.co/9rolb07wOy
20
16
66
this is gonna change the engineering field completely just knowing how to code isn't enough anymore, and the engineering manager's role will go away staff engs will make 10x the impact understanding product, business, and systems is more critical than ever
today we are introducing codex. it is a software engineering agent that runs in the cloud and does tasks for you, like writing a new feature of fixing a bug. you can run many tasks in parallel.
2
0
20
We're constantly experimenting with crypto AI agents - this is just the start! More exciting updates coming soon. :) Shoutout to @cjberragan for shipping this!
1
0
7
The result? A huge improvement from the previous architecture Most onchain actions now complete with just 1-2 inference calls As foundational models get better at coding, Nebula will become smarter and more capable of handling complex tasks
1
0
6
In the new architecture, Nebula generates and executes code snippets to complete tasks. This is where the magic happens: It thinks in code. The snippets may use one or multiple tools, reducing the number of inference round-trips
1
0
7
Context: Nebula typically uses 3+ tools to perform basic onchain actions -- understanding chains, researching contracts/tokens, then executing Each tool call is a model inference (slow + costly), and outputs are interdependent. That means we can’t just merge them into a single
1
0
6
Engine Cloud is the simplest, most powerful tool in the thirdweb suite. → non custodial smart server wallets → auto funding, no gas needed → simple HTTP interface → handles nonces, retries, monitoring → webhooks for every tx and takes 3 clicks to get started for free 🤯
3
11
27