@techwithramesh
Account based inSri Lanka
About this account
- Account based in
- Sri Lanka
- Connected via
- Sri Lanka Android App
Account-level information from X, not a live location or the device used for a specific post.
CTO of Fcode Labs and Forevate
Sri Lanka
Joined April 2014
- Tweets672
- Following605
- Followers102
- Likes2.4K
I think this is going to be massive.
Have seen people try computer-use with Jev and it is very fast.
Curious about what @trycua can offer.
1/ Fast Computer Use is now solved with @typesafeai Jev + Cua Driver.
Available in development preview for macOS, Windows, and Linux. We call it jev-use.
Draft #3943: github.com/trycua/cua
𝗔𝗴𝗲𝗻𝘁 = 𝗵𝗮𝗿𝗻𝗲𝘀𝘀 + 𝗺𝗼𝗱𝗲𝗹
You rent the model every month; the harness is the part you can actually own.
Pick one you control, learn it deeply, and stop shopping.
Fluency compounds.
Oh, that was quick..
Jev by @typesafeai is now on OpenRouter, in beta.
Jev is a System One model. Instead of generating text, it takes your app's state plus a typed question and returns a typed decision with a probability attached. There is no JSON prompting, parsing layer, and nothing to validate against.
Thanks @typesafeai for the access! Just logged in and went through the walkthrough.
Let's see what I can build with it👀
A green check proves an assertion has matched. Which decision the evidence authorizes is a separate question, and it deserves its own answer.
A test result is input to a release decision - 𝗶𝘁 𝗶𝘀 𝗻𝗼𝘁 𝘁𝗵𝗲 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻.
Jev is taken over my feed now.
it's been < 36 hours
we've jev-ed 140k off the waitlist (our platform team trying hard to get EVERYONE off ASAP)
I've been told we're 9th biggest launch video of the year (and in very good company)
SO many cool demos! evals! ideas!
and still the most important number is going to be how much true automation y'all deploy!
If the design document begins with a scale you don't have, it's not truly architecture. A straightforward, shippable solution is preferable over a distributed and theoretical one every time.
New labels, old parts: harness, loop, graph, orbs sit on control loops, state machines, test harnesses, and per-job VMs. Learn the mechanics. Treat the vocabulary like weather.
A new AI term earns its name when the job changes.
Context engineering did. The model still sees a prompt. Choosing what to retrieve, what to summarize, and how tools return results are different jobs.
Older names for harness, loop, graph, and orbs:
techwithramesh.com/blog/we-a…
Own your AI coding harness so that a price rise means switching models and a config change through a router. Let the vendor own it, and the rise means relearning how you work.
What happened to Gemini CLI:
techwithramesh.com/blog/open…
Seen a model swap take more than a config edit?
ALT Own the harness; Rent the model.
they just killed like 1000 startups
Now available: ChatGPT for Financial Services.
This is a tailored ChatGPT Work experience that combines built-in financial data with GPT-6 Astra’s reasoning.
Teams can develop research, build financial models, and create customized client materials.
openai.com/index/introducing…
You can't check a fix for "make the export faster" yet. Faster than what? Measured how?
Half of breaking down a task is spotting that the requirement is undefined, then asking instead of guessing.
The other four skills worth a junior's time:
techwithramesh.com/blog/juni…
Two-factor auth and short-lived tokens answer "how do we stop this technically." They don't answer "how do we stop someone from being talked into handing it over anyway", and that's the attack that keeps working.
Ramesh Rathnayake retweeted
🚀 Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient.
🔹 Introducing the smallest model in our new architecture family, with native visual understanding.
🔹 Designed for greater capability, faster inference, higher throughput, and scaling to larger models.
1/6
Every "agentic AI" launch post skips the same sentence: giving a model tool access doesn't just make it faster. It makes its mistakes autonomous.
Retrieval that returns something related isn't retrieval that returns something useful. Ask a support bot about one invoice, get back a paragraph on invoicing best practices. That's not a bug, that's top-k similarity working exactly as designed.
A benchmark score is a claim about a test harness before it's a claim about a model. Swap the harness, watch the ranking move twenty points, and nobody reruns the marketing slide that already shipped.
Harness engineering: coined February 2026, called "becoming ever less important" by August.
Six months to learn a name, not a skill. The practices underneath are older than all of it.
techwithramesh.com/blog/we-a… #AIAgents
ALT Six months of harness engineering
