Roberto Allende retweeted
Naresh Jain compares coding agents with and without a harness. He found it made a frontier model more expensive in tokens for little gain, but make a weaker (and cheaper) model more effective.
martinfowler.com/articles/ar…
Roberto Allende retweeted
Atrophy is a choice. Machines play chess better than we do, but we still play and reap many benefits from it. Let AI perform tasks, but never stop challenging your mind, or yourself.
Roberto Allende retweeted
I’ve spent 30 years watching coaches try to develop intelligent players.
Here’s something I’ve learned:
▪️You cannot develop decision-makers by making every decision for them.
#TOVO
Roberto Allende retweeted
Jev + Claude Opus 5.5... i can't understand why everyone isn't wiring this up yet
the forks alone drop from about $110 a day to $4 once jev takes them over. right now it's the best way i know to run opus
what jev handles:
-> scores which project notes matter before the context even loads
-> hands the easy steps to a faster worker
-> picks a recovery path when a tool fails, so nothing retries blind
-> runs the focused checks before the full test suite
opus does the hard reasoning. jev chooses from options the harness already prepared and validated
that's jev engineering in one line: the expensive model thinks, the cheap one decides
$106 a day back on the same work is a number your manager remembers
read the paper, then the full jev guide below
Ollama now supports Jev-like decision models all locally.
Use decision models like Nimble for tasks like ticket triaging, model routing, and content moderation.
ollama pull nimble
Here’s Nimble playing Ollama racer through the new local /v1/systemone API by making decisions in real-time. 🏎️
Roberto Allende retweeted
Jev for RAG, clearly explained!
Hybrid search gives you a shortlist. It does not decide which passages contain evidence, which are merely adjacent, or whether the evidence is strong enough to answer.
That missing judgment is where Jev fits.
Jev sits between retrieval and generation. It does not replace BM25, embeddings, or the LLM. It evaluates the candidates they produce before those candidates enter the context window.
→ Retrieve wide
Combine dense and keyword search, then merge the results with reciprocal rank fusion. This gives Jev a broad candidate pool, such as the top 20 passages shown in the visual.
Retrieval still sets the ceiling. If the right passage is missing from this shortlist, Jev cannot recover it.
→ Judge every candidate together
Send the query as Jev’s state. For each candidate, ask a typed yes-or-no question such as “Does C7 help answer this query?”
Jev evaluates every question in one packed request and returns a calibrated probability for each passage. This avoids making a separate model call for every query-passage pair.
→ Let code apply the threshold
Your application compares each probability with a threshold. Candidates above it continue to the LLM. Everything below it is removed before generation.
Jev makes the fuzzy judgment. Code remains responsible for the actual decision.
→ Gate the entire answer
The same request can check whether the retained passages make the query answerable. It can also flag signs of prompt injection inside a candidate.
If answerability falls below the threshold, the application can skip the LLM and return “not in the documents.” The injection score should remain a filtering signal, not a security boundary.
The result is a cleaner division of work.
Hybrid search retrieves broadly. Jev reranks, filters, and decides whether sufficient evidence exists. The LLM writes only from the passages that survive.
Jev’s value here is not simply moving passages up or down a list. It turns relevance into an explicit probability that your application can inspect, threshold, and act on.
To summarise:
- Retrieval finds the candidates.
- Jev decides what deserves context.
- The LLM writes the grounded answer.
----
I also built an open-source project showing how to use Jev as a judge for AI observability with Comet Opik.
It evaluates support traces for groundedness, request coverage, action honesty, and helpfulness, then records the results as an auditable experiment.
You can explore the project here: github.com/patchy631/jev-as-…
My article on how Jev works is quoted below.
Roberto Allende retweeted
One of the biggest shifts in software engineering may be an identity shift:
From writing software to making software.
AI can take on the implementation.
But understanding the problem, making trade-offs, designing the system, spotting complexity, and owning the outcome still matter.
The question may NOT be: “How much code will AI write?”
It may be:
“What kind of software maker do I want to become?”
Roberto Allende retweeted
Economic freedom is the only surefire way that humans have discovered to create wealth. stevestewartwilliams.com/p/a…
Roberto Allende retweeted
YC W19's @HeartAerospace just flew the largest electric airplane ever flown.
Seven years ago, co-founder & CEO @AndersForslund1 came to YC with a 3D-printed model you could hold in your hand. Today, the aircraft has a 100-foot wingspan, weighs 25,000 pounds, and costs just $5 in electricity to get off the ground.
YC’s @gustaf visited their pilot plant in LA to learn how they built it -- and what cheap electric flight could unlock for the future of regional air travel.
🚨 David Beckham on Lionel Messi after his four-goal performance against Montreal:
🗣“I’m still in awe. I honestly don’t know how he continues to do things like this so effortlessly.
Before the game, we spoke about the team’s recent form. He told me he had spoken with the coach and the players about changing their mentality and getting back to winning ways.
That’s exactly what I saw from the team tonight.
He’s a true leader. He leads by example, but he also does so much behind the scenes to help the team. That’s what makes him so special.
I’ve been lucky enough to witness what he did at the World Cup, and now to see him doing this again. Everyone who watches him can appreciate just how special he is.
This generation is incredibly lucky to witness Lionel Messi.”
Don't do vibe coding. Micromanage
allende.nz/notes/fourteen-mo…
Fancy Kanban — a Kanban plugin built for AI Agents and Humans
forum.obsidian.md/t/fancy-ka…
60
Roberto Allende retweeted
🧠 Pep Guardiola breaks down football management into it's simplest form:
🗣️ Pep: "There are a lot of managers who are really good tactically. There are good managers when it comes to man-management. There are managers who are really good in the media.
But there is one characteristic they all need to survive in this world: you have to convince your players.
In the end, I just give the players the weapons to express themselves the best they can. And when you change shapes and systems, it’s to give them that feeling of, “Oh, we’ll feel more comfortable playing this way to win games.”
The rest is poetry."
New video course: Micromanaged Driven Development
I've been using this AI coding workflow since June 2025, across multiple projects. In this course, I build a full React app with Kiro CLI.
youtu.be/1uIAbS3Tsvc?si=e-2K…
1
88
Roberto Allende retweeted
Que hermoso esto que dice Manu Gutiérrez sobre Messi y como le dio lugar a entrevistarlo en el Mundial, lo que le dio mucho más reconocimiento al público ❤️🩹
Shipped MMDD v5.
Rewrote the core prompt as a compact AI instruction — half the tokens, all the control. Also added a drop-in kanban extension for tracking units.
Still pure Markdown.
Still tool-agnostic.
Still micromanaged.
mmdd.dev
72
Roberto Allende retweeted
Weekend Productivity Challenge: Chur List
#productivity #alexa #amazonbedrock
builder.aws.com/content/3Akl…


























