Blog

From the Forge

Thoughts on AI intelligence, learning systems, and building technology that compounds.

context-engineeringfounder-notesagentsdecision-gatesmemory

What Your AI Is Allowed to Write Down

Context engineering is a mature theory of what goes into a model's window. What comes out of a session and becomes permanent gets treated as a storage tactic, not as a decision with an owner. The compliance deadline everyone cites just moved by sixteen months, so the case has to stand on engineering grounds.

August 12, 2026

eu-ai-actcompliancereferenceagentsgovernance

The EU AI Act's High-Risk Deadline Moved to December 2027. Most Sources Still Say August 2026.

Regulation (EU) 2026/1744 entered into force on 27 July 2026 and deferred Chapter III high-risk obligations by sixteen months. The deferral is unconditional, which is the opposite of what the proposal said and of what much of the internet is still reporting. Here is what moved, what did not, and how to check it yourself.

August 7, 2026

context-engineeringfounder-notespersonalizationlearningsycophancy

The Memory That Agrees With You

Measured in 2026: adding a memory layer amplifies sycophancy up to 25x over the same model with the conversation already in context, and explicit memory improves recall without improving how well the answer fits you. The cause is what these systems learn from. They store what you said. They should be learning from what you kept.

August 7, 2026

mcpguidesprotocolbridgesengineering

MCP Went Stateless: Everything That Changed in the 2026-07-28 Spec

The Model Context Protocol removed the initialize handshake, protocol-level sessions, and the Mcp-Session-Id header. Every request now carries its own version and capabilities. Here is what changed, what breaks, and what to do about it.

July 29, 2026

mcpguidescross-llmbridgesportability

How to Connect a Custom MCP Server to Any LLM: The Complete Guide

Every AI app that supports custom MCP servers in August 2026 — Claude, ChatGPT, Gemini, Perplexity, Cursor, and 30+ more — with exact setup paths, plan requirements, the transport-naming trap, and how open models like GLM and Kimi get MCP through a harness.

July 18, 2026

founder-notesconvergencecompoundingwhy-now

Why Now

For two years the question in AI was which model you used. That race is ending as the models converge on capable. When the engine is common, the edge moves up a level, to the layer that carries how you think across every tool.

July 5, 2026

founder-notesmemorycontext-rotcompounding

Why Every AI Memory Tool Hits a Wall at Month Three

Most AI memory tools feel sharp in week two and generic by month three. That drift has a name now, context rot, and it traces back to what storage was built to collect. A system that learns instead of stores bends the curve the other way.

June 28, 2026

founder-notesportabilitycross-llmlaunchv1-launch

The Bridges Change

For a year I was a courier for my own memory, hand-carrying context between AI tools that refused to talk to each other. This is the layer I built so the part that's you travels while the models stay rented.

June 14, 2026

founder-notesoperator-storylaunchv1-launch

The Operator's Operating System

I was hired company after company to fix 'a sales problem.' It almost never was. The judgment you build finding the real one is yours, and Tempreon is the layer that keeps it.

June 7, 2026

founder-noteslaunchv1-launchthe-journey-is-human

The Journey Is Human

After a year of trying to make AI work the way I do, I learned that the journey is the part nobody else can carry for you. Tempreon is the layer that does.

May 26, 2026

launchproductvision

Welcome to Tempreon

Introducing the intelligence forge — an AI layer that learns who you are and gets smarter with every interaction.

April 6, 2026