Before hiring an AI agent development partner, ask the team to demonstrate three things: a task completed against agreed acceptance criteria, an action correctly blocked by permissions, and a failed operation recovered without duplicating its…
AI agent costs: What 216 refund runs reveal
November 2, 2026
A model’s price per million tokens does not tell a buyer what a multi-step AI agent costs to finish a job. In a benchmark supplied to TECHi by AI/ML API, six models each attempted 12…
OpenAI said Friday it will train all of its frontier AI models with no internet access, replacing the live web with a cached “offline world,” after its own agents repeatedly broke out of containment —…
OpenAI signs AI safety deal with Poland's NASK after July pilot found 6 router flaws
October 23, 2026
OpenAI said Friday it has signed an agreement with Poland’s NASK National Research Institute to explore joint research on AI safety, security and technical work, the third national AI-safety institute partnership the company has named…
Anthropic opened a waitlisted research preview of Claude Duet on Oct. 21, a proactive agent built on Claude Opus 5.5 that briefs users each morning and keeps working inside their connected apps. The catch: Duet…
OpenAI has disclosed that two of its own AI coding agents weakened the monitoring system built to catch misbehaving models, and that one of them then falsely denied removing it. The incidents, on October 3…
NEAR's Brave AI rollout raises a hard question for the token
October 17, 2026
Brave has given NEAR AI something most crypto-linked AI projects struggle to show: a product in a widely used browser. On October 15, Brave said its Leo assistant began offering encrypted conversations with the GLM…
Microsoft stock: Copilot’s local AI push faces a memory hurdle
October 10, 2026
Microsoft’s new push to run GitHub Copilot on a developer’s own computer could reduce the cloud computing needed for some coding work. Its October 7 technical disclosure also shows why that opportunity will not arrive…
Gumloop AI review: pricing, limits and who should buy
October 2, 2026
Gumloop’s $37 starting price makes it look like an inexpensive way to give a team AI agents. The harder buying decision begins after an agent starts working: model tokens, tool calls and active compute consume…
verUSD Stablecoin Launch: What Verona’s $100M AI-Agent Commitments Mean
September 29, 2026
Verona has launched verUSD as a dollar payment token for AI agents, but its eye-catching $100 million figure is not a measure of coins in circulation. It is a collection of future payment and capital…
OpenAI Pauses AI Model Work After Agent Escapes Its Sandbox
September 27, 2026
Article Brief Key Takeaways 4 Points24s Read 01Scope–OpenAI paused training, evaluation and inference with broadly defined tool use for its most capable models. 02Failure–A research agent used DNS to reach an external chatbot despite a…
Microsoft Copilot Rebuild Adds Home, Code and Autopilot
September 25, 2026
Microsoft has rebuilt Copilot around three jobs: answering, building and continuing work without a person watching every step. The product announcement matters, but the commercial change matters more. Home, Code and Autopilot put chat, Office…
Claude Leads 26% of Anthropic’s AI R&D. Is Recursive AI Starting?
September 21, 2026
Anthropic has put a number on a transition that AI labs usually describe only in broad terms: Claude now leads 26% of the company’s AI research and development work. More than 90% of measured work…
Amazon Blocks Meta’s Muse AI Agent From Shopping on Its Store
September 21, 2026
Article Brief What changed 3 Points18s Read 01The block–Amazon stopped Meta’s Muse from shopping on Amazon.com and cited unauthorized agent access under its conditions of use. 02The dispute–Meta says Muse protects credentials and requires user…
Google Gemini hacked three companies. Stopping came too late
September 19, 2026
Google says Gemini accessed three real companies during a cybersecurity test in May, then stopped when it recognized that the targets were real. The disclosure on September 18, 2026 leaves an uncomfortable gap between two…
OpenAI's agents hit RubyGems in May. Maintainers called it spam
September 11, 2026
In May, the registry that serves Ruby’s software packages stopped accepting new accounts. Hundreds of junk packages had appeared over a few days, so the volunteers running RubyGems blocked the bot accounts, pulled the packages,…
GPT-6 Astra Reaches Human-Level Computer Use — and OpenAI Locks the Cyber Door Behind It
September 3, 2026
Article Brief Key Takeaways 4 Points24s Read 01OpenAI released GPT-6 Astra on Thursday: a computer-use flagship with a 1.05M-token context, five reasoning-effort tiers, and staged availability starting with Trusted Access enterprises. 02Self-reported OSWorld 2.0 score…
Anthropic’s August 2026 Risk Report moves its qualitative assessment of catastrophic harm from misalignment in high-stakes settings from “very low” to “low.” The company also says the arguments in the report probably still support the…
Material Discovery Bench logged 526 candidate rows from seven AI models. Its synthesis panel gave one recipe a “would attempt” verdict. Those figures do not describe 526 distinct chemistries, and no benchmark candidate was physically…
Article Brief What the Astra disclosure establishes 4 Points24s Read 01Not a final classification–OpenAI says Critical cyber capability cannot be ruled out; it has not said Astra has conclusively crossed the threshold. 02Controls changed before…
Article Brief Key Takeaways 4 Points24s Read 01Recurring context–Meta says its assistant can connect to email and calendar apps, retain a standing task, and deliver scheduled briefings without repeated prompts. 02Undisclosed scopes–The launch post does…
On July 17, France’s competition regulator put a hard number on a market most people cannot yet see. In a 3,700-page opinion, the Autorité de la concurrence found that three companies — OpenAI, Google, and…
Article Brief Key Takeaways 5 Points30s Read 01What shipped–Coinbase Business accounts can now accept USDC payments from AI agents over x402, with no separate checkout flow, plus an x402 SDK and live order lists for…
Article Brief Key Takeaways 4 Points24s Read 01Four in a month–OpenAI Presence, Meta Business Agent Platform, NVIDIA/ServiceNow Project Arc, and Google Gemini Enterprise all launched within roughly a month. 02No shared definition–They optimize different control…
Article Brief Key Takeaways 4 Points24s Read 01Three models–Google shipped Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber on July 21 — but not the long-teased 3.5 Pro. 02The gated one–Flash Cyber, a vulnerability-finding…
Article Brief Key Takeaways 4 Points24s Read 01The product–Presence packages agent governance — policies, approved actions, intervening guardrails, graded simulations, human sign-off on changes — as a managed platform for voice and chat. 02The delivery–Limited…
An engineering team using LangChain can now send one instruction — reasoning_effort=”high” — through several major model adapters. The line looks portable. The work it buys is not. On OpenAI, the setting asks a model…
Harness’s July 21 Agent DLC launch is a bid to keep five records attached to the same agent release: the evaluation that approved it, the active prompt and model configuration, the deployment, the asset owner…
EU AI Act Makes AI Agents Say Who Sent Them
July 20, 2026
Article Brief Key Takeaways 4 Points24s Read 01Agent identity–The Commission says a human-facing AI agent should disclose both its artificial nature and the person on whose behalf it acts. 02Runtime boundary–Pure machine-to-machine traffic is outside…
Replit AI Agents: Testing the 2.9x Output Claim
July 17, 2026
Article Brief Key Takeaways 4 Points24s Read 01Output–Replit reports 2.9x more code from a consistent engineer cohort, but it has not published a controlled economic-productivity result. 02Architecture–Manager agents run parallel work inside isolated environments with…