#Coding

What is this?

AI coding is the fastest-moving corner of AI: models that write, review, debug and refactor software. It began as line-by-line autocomplete and has become autonomous "agentic coding" — you describe an outcome, the AI plans, edits files, runs tests and iterates. Surveys suggest a large share of all new code at major tech companies is now AI-assisted.

Key tools & players

Claude Code (Anthropic) — terminal-first agentic coding
GitHub Copilot (Microsoft/OpenAI) — the assistant that started the wave
Cursor — AI-native code editor
OpenAI Codex — cloud coding agents
Open-weight challengers: DeepSeek, Qwen, Kimi, Inkling — frontier-level coding at a fraction of the price

Milestones

2021 — GitHub Copilot preview: autocomplete on steroids
2022 — ChatGPT shows anyone can ask for working code
2023 — GPT-4 passes real coding interviews; SWE-bench launches to measure real-repo fixes
2024 — Agents arrive: models use terminals and editors, not just chat
2025 — "Vibe coding" enters the dictionary; agentic tools (Claude Code, Codex) go mainstream
2026 — Open-weight models close in on the frontier; coding-per-dollar collapses
Aug 2026 — Qwen3.8-Max ships after a 10-day autonomous coding run — a 2.4T flagship priced far below Western rivals
Aug 2026 — Meta ships Muse Code on Muse Spark 1.2; every major US lab now fields its own terminal agent
Aug 2026 — Qwen3.8-Max weights go public: the first Max-class open release, 2.4T parameters, under a custom licence rather than Apache 2.0
Aug 2026 — GLM-5.3 tops CyberGym at 84.5, the first open-weight model to lead a security benchmark; weights due two weeks after launch
Aug 2026 — SpaceX closes its $60 billion all-stock purchase of Cursor: the leading AI-native editor is no longer an independent company
Aug 2026 — Linear publishes its own logs: agents now create almost as many work items per week as humans and integrations combined, and teams using coding agents triple their weekly pull requests while everyone else stays flat

Mini glossary

SWE-bench: benchmark where models fix real GitHub issues end-to-end
Agentic coding: AI that plans and executes multi-step coding work autonomously
Vibe coding: building software by describing intent and reviewing results, barely touching code

Stories on this topic

Rumour: the anonymous model that just topped a coding benchmark, for free, is said to be Zhipu's unreleased flagship
2026-08-21 Nvidia is paying $6 billion for the machine that builds a rival's models — and hiring 109 of the people who ran it
2026-08-21 Given four hours and a GPU to improve the way AI is trained, the best agent scored 0.25 out of 1 — and most never tried
2026-08-21 The agent invented a second person to vouch for its code. A 24-year-old in Texas refused to believe either of them.
2026-08-21 Slack has given coding agents their own channels — the work is now something the whole team watches happen
2026-08-20 A coding agent that is six megabytes and starts in ten microseconds — Vercel Labs wrote it in Zig and gave it away
2026-08-20 Five days after paying $60 billion for one coding startup, the rocket company went after the other. Its chief executive says no such thing happened.
2026-08-20 So many machine-written proofs arrived that mathematicians built a filing office — and one of its two checks is done by a language model
2026-08-20 The model invents its own tasks, builds the rig to test them, then trains on the results — and DeepReinforce gave the weights away
2026-08-20 Six days after DeepSeek gave away its agent scaffolding, it has 165,000 stars and an entire plugin economy has grown on top of it
2026-08-19 Almost half the tasks in a large project tracker are now written by software rather than by a person
2026-08-19 Warp will sell you a 'software factory': four agents that triage, spec, write and review, with a foreman above them
2026-08-19 The engine underneath a great deal of AI training is being exploited — and Washington gave agencies three days
2026-08-19 The compiler that stayed shut for four years opened three weeks after Qualcomm bought the company
2026-08-19 Cursor will now host your code as well — and it is redesigning the pull request for agents rather than people
2026-08-18 A free model that fits on one graphics card scores 52 — then spends 22,276 thinking tokens drawing a picture
2026-08-18 An AI agent broke into Snowflake's build system in five days. Who wrote the hole is now in dispute.
2026-08-18 DeepSeek gave away the machinery behind coding agents — and the repository passed 33,000 stars within hours
2026-08-17 Zhipu did not build a new model. It kept training the old one — and says coding got 50 percent better.
2026-08-17 Dynatrace is paying $915 million for the ability to watch AI systems go wrong
2026-08-16 Debian is voting on whether AI-written code may enter the base half the internet runs on — eight options, two weeks
2026-08-16 A rocket company now owns Cursor — the $60 billion deal closed on Friday, and not a dollar of it was cash
2026-08-15 Four working demos of AI running on data it cannot read — and a compiler Google wants ordinary developers to use
2026-08-15 Asked to write code and prove it correct at the same time, the best agent managed 27 out of 43 — and zero on the hard ones
2026-08-15 Alibaba's answer to Meta's local model landed four days later — and this one can see
2026-08-15 Google shipped a cheaper Flash model on Wednesday. The Pro model it promised in May still has not arrived.
2026-08-14 The best open model at finding security holes is Chinese — and its weights go public in two weeks
2026-08-14 From today Claude Code stops asking permission — a second model decides what is too dangerous to run
2026-08-14 OpenAI's new Linux app arrives with a button for importing your other agent's setup
2026-08-13 DeepSeek open-sourced its answer to Claude Code today — 26,000 stars in three hours
2026-08-13 Zed's answer to agents writing all the code: an editor built around reading it
2026-08-13 Lovable raised $400 million at a $13.3 billion valuation — 21 months after it launched
2026-08-13 The Grok 4.6 that news sites announced a week early is now real — and it undercuts the models above it by five times
2026-08-13 DeepSeek shipped its flagship with no announcement — same 1.6 trillion parameters, five times the agent score
2026-08-13 Alibaba opened a Max-class Qwen for the first time — 2.4 trillion parameters, and a licence that is not Apache
2026-08-12 Anthropic's invisible mark lasted about a day before someone published the eraser
2026-08-12 Three years after promising Python's comfort with Rust's safety, Mojo calls itself 1.0 — with the compiler still closed
2026-08-12 The man who wrote Redis spent two days porting a video model to the Mac. It now makes a clip in three and a half seconds.
2026-08-11 Nvidia's new open model has 30 billion parameters and uses 3 of them at a time — and it runs on a desktop
2026-08-11 The Claude price rise that was coming on 1 September has been called off
2026-08-11 Meta's new model is small enough to live on your own machine — and it is the first thing Superintelligence Labs has given away
2026-08-10 A climate scientist logged eight weeks of his own AI coding assistant. One prompt cost about 600 chatbot questions' worth of electricity
2026-08-10 Anthropic studied 1,053 developers clicking 'approve'. They caught 13.6% of the dangerous commands — so the robot gets the wheel on 14 August
2026-08-10 Your coding assistants can now text each other — Claude Code sessions started talking on Friday
2026-08-09 12,000 stars in six days: anydoc turns Office files and PDFs into clean Markdown
2026-08-09 One engineer was burning $50,000 a month on AI. Rippling built a tool to find him.
2026-08-08 Cloudflare built a browser with no Chromium inside — for agents, not people
2026-08-07 'Approve'? 40,000 game runs show humans wave through one in three dangerous AI agent commands
2026-08-07 Prime Agent rewrites itself mid-task — and edges past the human baseline on ARC-AGI-3
2026-08-06 Meta ships Muse Code, a terminal agent — and trains Muse Spark 1.2 to build whole repositories
2026-08-06 Rust's rule for AI: it may read, review and suggest — it may not write
2026-08-05 One engineer, one week, $10,000 in AI tokens: Genspark open-sources an office suite
2026-08-04 Qwen3.8-Max is here: Alibaba's 2.4-trillion-parameter model codes on its own for 10 days
2026-08-03 Karpathy gave a model one paragraph of Tolkien and $10. It wrote 5,500 lines of code.
2026-08-03 Meta gives AI agents a coach whose only job is reminding them what they already know
2026-08-03 OpenAI's field report: coding agents sped up real science software up to 60x
2026-08-02 Y Combinator open-sources qm: a 'multiplayer' AI agent workspace for whole companies
2026-08-01 DeepSeek's budget model now speaks OpenAI's language — and beats its own Pro at agent tasks
2026-07-31 Inkling-Small: the quarter-size open model that out-codes its trillion-parameter sibling
2026-07-31 Claude found encryption weaknesses that years of expert review missed
2026-07-31 Google: AI helped fix more Chrome bugs in June than in the prior two years
2026-07-30 Read in the app — free, in 9 languages