AI Workflows & Agentic Architecture Directory

NVIDIA reported a $500 fine-tune of a 9B open model outperforming frontier AI on catalog review tasks. Here is the unit-economics breakdown of when open-weight fine-tunes beat frontier APIs — and when they do not.

Ten major AI agents support custom remote MCP servers with native OAuth 2.1, and the 2026-07-28 spec made MCP stateless and cacheable. Here is the state of MCP in 2026 — and what security must catch up on.

The EU AI Act's high-risk obligations took full effect August 2, 2026. Here is the practical compliance stack for agentic AI systems: classification, testing, logging, human oversight, and the audit trail that proves it.

METR's time-horizon measurements track how long frontier models can work without human help. Here is what the 2026 numbers mean for agent autonomy, deployment risk, and the oversight controls that have to keep pace.

LLM Ops in 2026 is production engineering, not prompt plumbing: eval harnesses, OpenTelemetry observability, model routing, caching and canary rollouts are the standard stack. Here is the maturity model and the tooling that gets you there.

L&T Technology Services unveiled AgenticIQ on August 11, 2026 — an end-to-end agentic AI platform for engineering and manufacturing. Its planning-first, reuse-by-design architecture is the pattern every engineering org will copy this year.

An August 2026 arXiv paper, 'Not an A11y,' shows Android accessibility — the privileged channel mobile GUI agents depend on — is also their injection surface. The same trusted path that lets them see screens lets untrusted apps hijack them.

For two years, the most useful thing an AI assistant could do was also the one thing it refused: pick up the phone and call a business for you. In August 2026 that gate finally opened — Google, Apple, and a wave of voice startups converged on the same conclusion.

On July 30, 2026 OpenAI cut GPT-5.6 Luna API pricing by 80% to $0.20 per million input tokens, spread a mid-tier drop across the lineup, and left the flagship price untouched. The model market has formally split into three tiers. Here is how to buy them.

A retrained small model just outscored its own larger flagship — without changing a single parameter. DeepSeek V4 Flash 0731 moved the Nasdaq, compressed the US-China gap to single digits, and reset what teams should pay for coding agents.

On August 8, 2026 OpenAI halted development of Astra — its next frontier model — after internal testing showed it could autonomously develop zero-day exploits and complete end-to-end cyberattacks. The same system solved ten long-unsolved math problems. The pause is a landmark: the first voluntary stop of the agent era over cyber capability.

AI agents don't log in like people. With non-human identities outnumbering humans 45:1 and 40% of live MCP servers running without authentication, teams need an identity control plane — not another spreadsheet.