TypeSafe CEO Diogo Almeida on Jev Model, System-One Architecture, and AI Automation Gap

·8 views

TypeSafe CEO Diogo Almeida discussed the launch of Jev, the company's new AI model, in a wide-ranging podcast interview, describing it as a 'System-One' programmable model optimized for intelligence per dollar rather than human-facing chat. He argued that mainstream RLHF training causes mode collapse, making models overly conservative and poorly calibrated for programmatic use cases. Almeida criticized embedding safety refusals at the API layer, comparing a reliable AI model to a database that should not unpredictably reject downstream developer calls. He emphasized that model capability stems primarily from carefully curated data rather than raw compute, and that TypeSafe's proprietary RLCD training objective is designed to remove humans from the automation loop. Almeida expressed confidence that Jev's release marks a turning point for programmable AI, noting the model entered real production workloads within its first week and that Discord community membership has already reached 100,000 users.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why Scaling MCP Beyond Local Dev Creates Serious Enterprise Governance Challenges

The Model Context Protocol (MCP) allows developers to connect AI models to tools and data quickly, making it popular for local development workflows. However, when organizations attempt to roll out MCP at enterprise scale, significant problems emerge around security, visibility, and governance. Hardcoded credentials, untracked server instances, and overly broad permissions granted to AI agents can expose organizations to serious risks. Platform and security teams struggle to audit which users, models, and tools are involved in any given operation when MCP servers are managed independently across teams. Experts argue that enterprise-scale MCP deployments require a dedicated gateway layer with centralized identity management, observability, and governance controls.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer builds cryptographic verification system to audit AI agent delegation

A solo developer has released an open-source MVP called AI-tower that makes AI agent delegation cryptographically verifiable using Ed25519 digital signatures. The system addresses a gap in current AI agent infrastructure, where monitoring alone cannot prove which agent authorized a specific action or capability handoff. When one AI agent delegates tasks to another, the system creates a signed payload recording who delegated, to whom, what capabilities were granted, and when. Verification is performed client-side using the Web Crypto API, meaning the server cannot tamper with delegation records after the fact. The project is available on GitHub, and the developer is seeking community feedback on the canonicalization approach and capability model.

0 comments Read more at DEV Community

ProgrammingHacker News ·

Robin Williams' Daughter Calls Out Fans Making AI Videos of Late Actor

Zelda Williams, daughter of the late comedian Robin Williams, has publicly criticized fans who are creating AI-generated videos featuring her father. She urged those responsible to reconsider their actions, telling them to 'have some shame.' The issue highlights growing concerns around the use of artificial intelligence to recreate deceased celebrities without family consent. The story was reported by Variety in 2026 and drew attention on social platforms and tech forums.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

PonyMux 0.10 Adds Full-Text Search Across Claude Code and Codex Chat History

AI coding agent tools like Claude Code and Codex store complete conversation logs locally on disk, but retrieving specific past discussions has long been difficult without manual grep commands or third-party utilities. PonyMux 0.10 addresses this by integrating full-text search directly into its existing Cmd+K terminal launcher, allowing users to search across all stored conversation transcripts from a single interface. The feature indexes every prompt, agent response, and tool call from local JSONL and session files, returning matched snippets with highlighted keywords. The index is built using SQLite FTS5 with trigram tokenization, enabling substring and CJK language matches without extra configuration, and updates automatically via filesystem events after the initial 10–30 second setup. The search index is stored separately and can be paused or rebuilt from settings without affecting other PonyMux data.

0 comments Read more at DEV Community