The spec dated July 28 drops protocol-level sessions entirely and starts a twelve-month countdown on Roots, Sampling, and Logging — the largest migration server authors have faced since the standard launched.
· 7 min · Tools
Opus 5 costs the same as the model it replaces, half what Claude Fable 5 costs, and outscores Fable 5 on agentic coding — while still trailing it by 0.8 points on SWE-bench Pro.
· 5 min · Tools
Moonshot's 2.8-trillion-parameter model is the largest ever released with open weights, and it's beating Fable 5 on an independent coding leaderboard — but it dropped the cheap-Chinese-model playbook, and the weights it's named for aren't downloadable yet.
· 5 min · Tools
The IDE-native agent hit general availability topping an independent benchmark at a quarter the cost of its nearest rival — by doing something terminal agents structurally can't: driving the real debugger.
· 4 min · Tools
Three separate 2026 findings — the GuardFall shell-injection class, a poisoned Claude Code GitHub Action, and hallucinated packages installed on autopilot — point at the same thing: the agent with a shell is the new attack surface.
· 5 min · Tools
Windsurf quietly became Devin Desktop, and on July 1 the on-machine agent it ran for two years went dark — a zero-effort switch if you code by hand, a broken build for anyone whose CI still calls Cascade by name.
· 4 min · Tools
A controlled study clocked experienced developers 19% slower with AI while they swore it had sped them up — and three more datasets show the gap isn't just a feeling.
· 5 min · Tools
Anthropic finally set a date to fold its best model back into Max and Team Premium at no extra charge. Pro subscribers get left out with a one-time $100 credit, and even the included plans see the cap land at half of a limit that shrinks a third the same day.
· 6 min · Tools
The same week GitHub opened its desktop app and a second AI agent to every plan tier, it also shipped new caps letting admins stop that wider access from quietly blowing through a paying team's budget.
· 3 min · Tools
Entire's distributed Git network claims triple the clone throughput of Cursor's rival Origin platform — though both companies' numbers come from their own demos, not an independent bake-off.
· 3 min · Tools
Kiro decided rebuilding AWS Cost Explorer from scratch was the fastest bug fix, and did it without a pause for approval. Two much larger AI-linked outages followed months later, and Amazon's public account still calls it human error.
· 4 min · Tools
After years of giving Llama away for free, Meta is charging $1.25 and $4.25 per million tokens for a model that tops the field on workplace tool use and trails it on raw coding accuracy.
· 4 min · Tools
xAI's first purpose-built coding model clears an agentic benchmark that specifically penalizes rule-breaking at a third of Fable 5's per-task cost — but breaks more rules getting there.
· 4 min · Tools
A model that identifies a fact correctly 99.3% of the time in a short prompt drops to 69.7% on the identical task once the prompt runs a few tens of thousands of tokens — nowhere near its advertised limit. Most assumptions about how context and memory actually behave don't survive contact with the research.
· 5 min · Tools
Claude Code's own documentation splits 'keep working without me' into three distinct mechanisms — and the one nicknamed after a Simpsons character isn't the one built for checking on a deploy overnight.
· 5 min · Tools
Anthropic gave its office-work agent a phone and a browser this week. The more telling news is buried in its own usage data — only 8.7% of Cowork sessions are software development.
· 4 min · Tools
Intent doesn't write code faster than a single coding agent. It exists because tracking which of five agents is doing what became the actual bottleneck.
· 3 min · Tools
A survey of 2,847 developers found reviewing AI-generated code now eats more hours per week than writing it — and independent telemetry data backs up the shift.
· 4 min · Tools
Sol, Terra, and Luna beat every OpenAI model on coding and cybersecurity benchmarks. Before any paying customer could touch them, the Trump administration got a preview and a say in the rollout.
· 3 min · Tools
Anthropic gave every Pro, Max, and Team subscriber six days of included access to its best model. Today the free ride ends, and what happens next depends entirely on which plan you're on.
· 4 min · Tools
Fable 5 tops Anthropic's own coding benchmark by 11 points and ships to every paid plan. Its twin, built from identical weights, is restricted to about 50 organizations hunting zero-days that survived 27 years in critical software.
· 3 min · Tools
DeepSeek's newest cheap coding model claims to match Claude Sonnet 4.6 on the industry's most-cited benchmark at roughly 1/20th the price. Only one of those two scores has been independently checked.
· 5 min · Tools
It calls itself voluntary. Anthropic cooperated anyway, and ten days after Claude Fable 5 shipped, the government hit it with export controls over a bug the company called minor and industry-wide.
· 4 min · Society
Both are legitimate, both are growing fast, and neither number in that title tells you which one belongs on your machine. Here's the framework that actually does.
· 6 min · Tools
Anthropic quietly gutted /agents's interactive wizard on July 1 — the third native command stripped down this way in two months. Here are the ten built-ins still worth keeping in your fingers, ranked by how much they actually change a session.
· 5 min · Tools
84% of developers now use AI tools. The 2025 Stack Overflow and JetBrains surveys show they're being far more selective about which tasks they'll actually trust it with than the adoption number suggests.
· 4 min · Industry
Most community skills are marginal — a handful genuinely change how a session goes. Here are five that keep showing up across the best collections in 2026.
· 5 min · Tools
"MCP is for agents, APIs are for developers" is true and mostly useless. Here's the actual framework, with real token costs attached.
· 5 min · Tools
Claude Science quietly launched the same week Anthropic hired away DeepMind's Nobel laureate. Neither is a coincidence.
· 4 min · Industry
Three tools now call themselves agentic coding assistants. They don't mean the same thing by it — a side-by-side look at Antigravity 2.0, Claude Code, and Codex.
· 5 min · Tools