Juanchi.dev
Digital Native · real systems
·
HomeCVBlogLabContact
Juanchi.dev
Digital Native · real systems
·
HomeCVBlogLabContact

Agents and LLMs with judgment

39 posts on LLMs and 30 on agents: sandboxes, guardrails, cost per task, and running your own benchmarks instead of trusting screenshots.

  • Agents that execute

    Sandboxes, permissions, and what happens when an agent actually runs commands.

  • Measuring models

    Comparisons on your own cases, with cost and latency, not generic leaderboards.

  • The real cost of AI

    Tokens per task, caching, and where the monthly bill comes from.

Recommended reads

View more
Aug 17 2026

Noroboto: Lying Fonts and Rust Mitigation — A Technical Read Without the Hype

Fonts lie. Noroboto documents how the text subsystem can return incorrect metrics and proposes mitigations in Rust. Before you copy it into production, you need to understand what problem it actually solves, where the common recipe breaks down, and what reproducible experiment is actually worth runn

Aug 17 2026 · 8′ · Opinion · linux · sistemas

8′
Aug 17 2026

Cline in production: the autonomous code agent for VS Code I use with deliberate constraints

Cline can create files, run commands, and open the browser autonomously from inside VS Code. That sounds like productivity. It also smells like risk if you haven't thought through the permissions before you start. My thesis: the mental model matters more than the tool.

Aug 17 2026 · 9′ · Tutorials · TypeScript · LLM

9′
Aug 02 2026

Qwen3 locally with Ollama: what changed in the architecture and whether it's worth switching

Qwen3 landed with thinking mode and real improvements in code generation. But before you replace the model already running in your Ollama setup, there are technical questions you need to answer first. I answer them here without selling hype.

Aug 02 2026 · 9′ · Tutorials · TypeScript · Inferencia Local

9′
Jun 22 2026

Rate limiting in Next.js: what to protect before picking a library

Rate limiting isn't an npm dependency — it's an abuse policy. Before copying middleware, you need to define what asset you're protecting, what abuse pattern you expect, and what a false positive costs you. A guide with a decision matrix, real gotchas, and observability for Next.js.

Jun 22 2026 · 9′ · Tutorials · TypeScript · nextjs

9′
Jun 20 2026

OWASP LLM Top 10 in Production: How I Audited My TypeScript Agent Pipeline Against All 10 Risks — and What I Found

Running the OWASP LLM Top 10 as a real audit is a completely different experience than reading it as a checklist. I ran it against my TypeScript agent stack with system prompts, MCP tools, and Cline — and the findings were uncomfortable.

Jun 20 2026 · 9′ · Tutorials · TypeScript · LLM

9′
Jun 15 2026

Formal Methods and the Future of Programming: What's Worth Trying and Where the Ceiling Is

Formal methods keeps surfacing on the technical radar as the solution the industry ignored. My read: the problem it points to is real, but the recipe floating around omits costs that change the equation entirely.

Jun 15 2026 · 9′ · Opinion · TypeScript · sistemas distribuidos

9′

Juanchi.dev is Juan Torchia’s public notebook — architecture, real systems and product decisions. Written and reviewed by hand in Buenos Aires. Built with Next.js, Prisma and PostgreSQL; deployed on Railway. No third-party trackers beyond aggregate analytics.

BlogUsesNowGitHubRSS© 2026 Juan Torchia · All rights reserved.