The Jagged Frontier: When GPT-4 Helps Consultants and When It Does Not
Eighty-five percent of consultants working alone solved a hard business problem. Colleagues given GPT-4 were 19 percentage points less likely to get it right. On a different set of tasks, the same experiment found the opposite: higher quality, more speed, more work finished. The tool did not change. The job did. Those figures come from […]
Anthropic’s Agent Sandbox Trades Constant Approvals for a Configured Boundary
Anthropic’s agent sandbox trades constant approvals for a boundary you must actually configure Approve, not Run, is the most dangerous control in an AI agent’s day. It gets pressed for the fortieth time before lunch, long after anyone is reading the diff. Anthropic has been direct about that habit. Constant approve clicks slow the work, […]
Britain’s AI Budget Buys a Seat, Not the Keys
Shape the stack, or keep renting it There are two ways for a country to live with frontier AI. Rent it by the API call, or try to shape what gets built and how it is used. The Budget package now being trailed is an attempt at the second. It is smaller, patchier, and more […]
NVIDIA Nemotron 3.5 Bolt Follows Rules but Drops the Errand
The small model that follows the rules, then drops the errand A customer-service agent that knows when to stay quiet is useful. One that cannot finish the job is a liability. NVIDIA’s Nemotron 3.5 Bolt E4B 4GB currently looks like both. Bolt is a small reasoning model NVIDIA post-trained from Google’s Gemma 4 E4B instruction […]
Kimi K3.1 Pro: Cheap Tokens, Near-Opus Coding, No Weights Yet
Cheaper tokens, a coding chart that sits within a point of Claude Opus 5.5, and a system you still cannot run yourself. That is the offer in front of a buying committee. The useful question is which part of it your workflows can actually survive. Moonshot AI put Kimi K3.1 Pro on an API on […]
Multi-Turn AI Agent Training: Reward the Outcome, Not the Activity
In an early scoring scheme for a Wordle-playing agent, solving the puzzle on the third guess paid more than solving it on the first. The model was not being clever. The scoreboard was lying. Most leadership teams skip this part of agent training. They ask whether the model can write, classify, or call a tool. […]
Web Search for Coding Agents: Cheap, Fast, and Complete Don’t Align
On a public board that tests web search for coding agents, the cheapest provider is not the one that finishes the most work, and it is not the fastest. The sample is 100 tasks. Completion is the mean of three runs, not a large trial. TinyFish finishes 59.3% at a median cost of $0.035, with […]
Big Tech’s AI Spending Is a Payback Test, Not a Bubble Call
An analyst asked Microsoft, on an earnings call, the question much of the market has been muttering: “Are we in a bubble?” On Alphabet’s call, another asked what early signs showed the spending was producing better long-term returns. A technology-spending roundup dated 1 November 2025 recorded both. These were not posts from the cheap seats. […]
Bybit AI Adds Memory. The Advice Question Stays Unanswered
Wrong answers are the obvious failure. The expensive failure mode of an in-app assistant is a correct answer that forgets the customer who asked. Bybit wants Bybit AI to stop doing that. On October 22, 2026, the cryptocurrency exchange, which calls itself the New Financial Platform, announced personalization and memory features for Bybit AI, the […]
Unsourced AI Forecasts Do Not Belong in the Leadership Deck
A publisher can put a famous name next to a chatbot and call the result a forecast. That is what a CryptoNews page does with XRP. The headline is “Meta AI Predicts a Wild XRP Price: Does the Technical Analysis Align?” The body attributes the call to “Mark Zuckerberg’s Meta AI.” On the same page, […]