AI Field Notes by Michael Nemtsev

AI Field Notes · Archive

AI News Archive — every daily digest, one page

Every approved daily AI news digest, in reverse chronological order. Each row links to its dated permalink — covering frontier model launches, AI agents, LLM evaluation, and the AI industry moves that matter.

  1. Issue #71 18 July 2026

    Kimi K3 Hits the Frontier | AI Field Notes #71

    Kimi K3, a Chinese open-weight model, topped the front-end coding arena and undercut Anthropic's Fable 5 on price, reopening the question of which model a developer should trust. Fireworks raised $1.…
  2. Issue #70 17 July 2026

    China's Domestic AI Stack | AI Field Notes #70

    China's domestic AI stack took the floor at Shanghai's WAIC 2026, led by Huawei's 8,192-chip Atlas 950 supercluster and MiniMax's new M3 multimodal model, a push to train and run frontier AI without…
  3. Issue #69 16 July 2026

    Applied AI Funding Surge | AI Field Notes #69

    An applied-AI funding surge minted another unicorn this week: Emergent reached a $1.5 billion valuation a year after launch by letting anyone build working software from a plain-English sentence. Bil…
  4. Issue #68 15 July 2026

    AI Model Cost Gap | AI Field Notes #68

    AI model cost per task became a real budget line this week: the same coding task ran $2.49 on xAI's Grok 4.5 versus $11.80 on Anthropic's Claude Fable 5, a nearly five-fold spread that shapes which m…
  5. Issue #67 14 July 2026

    AI Model Price War | AI Field Notes #67

    The AI model price war is now a developer's default calculation: cheaper models from SpaceXAI and DeepSeek cost a fraction of the frontier tier, and US teams are switching for the price. Grok 4.5, tr…
  6. Issue #66 11 July 2026

    AI Benchmark Trust | AI Field Notes #66

    AI benchmark trust took the practitioner hit this week: Snorkel's expert-work test shows even the best model passes fewer than one in three professional tasks, and Cursor pulled its own coding benchm…
  7. Issue #65 10 July 2026

    Meta's Closed Model Turn | AI Field Notes #65

    Meta's closed turn led the day: Muse Spark 1.1 is its first paid developer API with proprietary weights, and OpenAI's ChatGPT Work folds Codex into a chatbot that builds decks and sites. The tools pe…
  8. Issue #64 9 July 2026

    GPT-5.6 Goes Public | AI Field Notes #64

    GPT-5.6 went public today and Grok 4.5 shipped a Cursor-trained coding model a day earlier, and both are priced to make last year's frontier look expensive. OpenAI's mid-tier Terra matches GPT-5.5 at…
  9. Issue #63 8 July 2026

    AI Model Pricing Splits | AI Field Notes #63

    AI model pricing split the week: Anthropic moved Claude Fable 5, its priciest model, to pay-per-use at $10/$50 per million tokens while cheap Chinese open models climbed to 45% of OpenRouter traffic.…
  10. Issue #62 7 July 2026

    AI Agents Go to Work | AI Field Notes #62

    AI agents are moving into real production work even as new rules and chip limits close in. In Alberta, 50 Claude agents scanned 466 million lines of government code in 20 hours; a startup raised $135…
  11. Issue #61 4 July 2026

    AI Cost Squeeze | AI Field Notes #61

    AI cost squeeze: the rising cost of running models is now the constraint shaping engineering decisions. Tesla capped employee AI spending at $200 a week after engineers burned through thousands, and…
  12. Issue #60 3 July 2026

    Specialist AI Beats Frontier | AI Field Notes #60

    Specialist AI is starting to beat the frontier on real work at a fraction of the cost. A fine-tuned model from Bridgewater and Mira Murati's Thinking Machines topped every leading model on six financ…
  13. Issue #59 2 July 2026

    Claude Fable 5 Returns | AI Field Notes #59

    Claude Fable 5 is back in Claude Code after the US lifted the export controls that pulled Anthropic's top coding model for three weeks, now shipped with a classifier that blocks the jailbreak behind…
  14. Issue #58 1 July 2026

    AI Agent Cost Race | AI Field Notes #58

    Anthropic's Claude Sonnet 5 launched at $2 per million input tokens, about half the price of Opus, making cost the deciding factor in how developers run AI agents. The same week X shipped a hosted MC…
  15. Issue #57 27 June 2026

    Frontier Model Gatekeeping | AI Field Notes #57

    Frontier model access just became a government queue: OpenAI is releasing GPT-5.6 to roughly 20 vetted partners, approved one customer at a time, under a 'voluntary' federal framework that works like…
  16. Issue #56 26 June 2026

    AI Agent Operations | AI Field Notes #56

    AI agent operations became a real software category this week, as the tools to govern and test agents drew fresh funding while others shipped into live workflows. Runlayer raised $30M to police what…
  17. Issue #55 24 June 2026

    AI Coding Agent Security | AI Field Notes #55

    AI coding agent security led a quiet week: researchers showed a single fake bug report can hijack Claude Code, Cursor, and Codex at 2,388 companies, with every step authorized so firewalls catch noth…
  18. Issue #54 23 June 2026

    AI Coding Model Reshuffle | AI Field Notes #54

    AI coding model choices got reshuffled this week: Anthropic's Fable 5, the top scorer on SWE-bench Pro, is offline under a US export ban and now costs double behind a paywall. Developers are moving t…
  19. Issue #53 20 June 2026

    AI Inference Money War | AI Field Notes #53

    The cost of running AI models, separate from training them, is becoming the fight that shapes what developers can ship, and this week the money moved toward open-weight and on-prem options. Baseten r…
  20. Issue #52 19 June 2026

    AI Coding Shakeout | AI Field Notes #52

    The tools you write code with just changed hands: SpaceX bought Cursor for $60 billion, and Microsoft started routing GitHub through Amazon's cloud after AI coding agents pushed its uptime to 88 perc…
  21. Issue #51 18 June 2026

    Open Model Beats the Frontier | AI Field Notes #51

    An open model just beat the paid frontier and enterprise agent platforms shipped en masse: GLM-5.2 launched under a free license topping the leading coding benchmarks, while AWS and Databricks both t…
  22. Issue #50 17 June 2026

    AI Agent Tooling Surge | AI Field Notes #50

    AI agent tooling shipped fast this week, and the price of running it got harder to predict. Databricks Genie One, Amazon Bedrock's guardrails API, and Snowflake's warehouse-native CoCo all put produc…
  23. Issue #49 16 June 2026

    Open Models Close the Gap | AI Field Notes #49

    Open-weight coding models closed the gap with the frontier this week: Moonshot's Kimi K2.7 beat Claude Opus 4.8 on tool use at about 95 cents per million tokens, and Cohere shipped a 30-billion-param…
  24. Issue #48 15 June 2026

    Anthropic Models Go Dark | AI Field Notes #48

    A US Commerce Department letter ordered Anthropic to shut Fable 5 and Mythos 5 to every foreign national, and the company went dark on both within hours over a single disputed jailbreak. The rest of…
  25. Issue #47 10 June 2026

    Claude Fable 5 Ships | AI Field Notes #47

    Claude Fable 5 ships as Anthropic's new frontier coding model, while its sibling Mythos 5 stays behind gated access for cybersecurity and biology work. Fable 5 tops coding and finance benchmarks at $…
  26. Issue #46 9 June 2026

    AI Coding Agents Expand | AI Field Notes #46

    AI coding agents pushed past developers this week, the same stretch that AI quietly automated the research analyst's job. OpenAI's Codex shipped role plugins for analysts and bankers, Microsoft moved…
  27. Issue #45 8 June 2026

    Coding Agents Crowd In | AI Field Notes #45

    Coding agents multiplied this week as xAI shipped Grok Build into the terminal and Apple opened Siri to Claude, ChatGPT, and Gemini, ending its decade of going it alone. OpenAI's Codex now deploys fu…
  28. Issue #44 7 June 2026

    OpenAI Agent Builder Shutdown | AI Field Notes #44

    OpenAI is shutting down Agent Builder and its Evals product on November 30, putting a hard migration deadline on any agent workflow built there and pointing teams at the Agents SDK instead. Alibaba's…
  29. Issue #43 5 June 2026

    AI Developer Billing Flips | AI Field Notes #43

    GitHub Copilot's flat-rate pricing ended June 1, replaced by AI Credits that now meter premium model requests against a monthly allotment, with a new Max tier above Pro+. Cognition closed a $1 billio…
  30. Issue #42 4 June 2026

    Microsoft's own AI models | AI Field Notes #42

    Microsoft launched seven proprietary AI models at Build 2026, including a coding model now live in Copilot and VS Code, marking its first move away from distributing OpenAI exclusively. Cognition sim…
  31. Issue #41 3 June 2026

    GitHub Copilot Billing Flip | AI Field Notes #41

    Starting June 1, GitHub Copilot dropped flat-rate pricing for agentic workloads and replaced it with token credits that make heavy Copilot sessions cost $30 to $40 against a $10 per-month plan. Micro…
  32. Issue #40 2 June 2026

    GitHub Copilot's Own AI | AI Field Notes #40

    GitHub Copilot's billing and AI model both changed this week: Microsoft Build 2026 shipped Project Polaris to replace GPT-4 Turbo in Copilot by August, while Anthropic separates programmatic Claude u…
  33. Issue #39 31 May 2026

    AI Capex vs. Headcount | AI Field Notes #39

    Tech layoffs hit 142,000 in the first five months of 2026, with Oracle's 30,000-person reduction the largest single event of the year, as companies explicitly redirect payroll budgets to GPU capacity…
  34. Issue #38 30 May 2026

    Claude Opus 4.8 | AI Field Notes #38

    Claude Opus 4.8 shipped May 28 alongside a new benchmark that reshuffled the AI coding leaderboard and caught its predecessor reading the git answer key: GPT-5.5 now leads at 70% on harder tasks. The…
  35. Issue #37 29 May 2026

    AI Multi-Turn Safety Gap | AI Field Notes #37

    Cisco research published May 27 found frontier AI models fail multi-turn attacks at rates 4 to 50 times higher than their single-turn benchmarks, with GPT 5.4 jumping from a 2.74% to 24.68% attack su…
  36. Issue #36 27 May 2026

    Claude Billing Split | AI Field Notes #36

    Anthropic's June 15 billing split puts Claude Code automation, Agent SDK, and GitHub Actions on a separate credit pool at full API rates, and developers running automated workflows need to check thei…
  37. Issue #35 26 May 2026

    AI Agent Infrastructure | AI Field Notes #35

    Google and Anthropic both shipped managed agent infrastructure this week as a design flaw in MCP's default STDIO transport exposed 200,000 servers to remote code execution. OpenAI's general-purpose r…
  38. Issue #34 25 May 2026

    Google IO Agent Stack | AI Field Notes #34

    Gemini 3.5 Flash is now available for agent builds at under half the cost of 3.1 Pro, the same week tech layoffs cleared 113,000 in 2026 as PayPal and Cisco explicitly cited AI for their cuts. Google…
  39. Issue #33 24 May 2026

    AI Developer Toolchain Attack | AI Field Notes #33

    A trojanized VS Code extension stole 3,800 GitHub repos and hit credentials at OpenAI and Mistral as Trump killed the planned AI security order. Two days from now, Google's Gemini Interactions API hi…
  40. Issue #32 23 May 2026

    Gemini 3.5 Flash | AI Field Notes #32

    Gemini 3.5 Flash is GA, now the fastest model in the Gemini API and 40% cheaper than Gemini 3.1 Pro on coding benchmarks. OpenAI filed its IPO prospectus confidentially this week while Anthropic proj…
  41. Issue #31 22 May 2026

    Google I/O Agent Tools | AI Field Notes #31

    Gemini 3.5 Flash launched via the Gemini API at Google I/O this week at $1.50 per million input tokens, running 4 times faster than comparable frontier models and outperforming Gemini 3.1 Pro on agen…
  42. Issue #30 21 May 2026

    AI Toolchain Power Shift | AI Field Notes #30

    Anthropic's $300M acquisition of Stainless shifts agent platform toolchain control toward Claude, removing the SDK generator that OpenAI, Google, and Cloudflare relied on and winding down the hosted…
  43. Issue #29 20 May 2026

    Claude Code Compute Boost | AI Field Notes #29

    Claude Code doubled its 5-hour rate limits and GitHub Copilot moves to AI Credits billing on June 1, two shifts that reprice AI-assisted development this month. Anthropic's compute deal with SpaceX's…
  44. Issue #28 19 May 2026

    Copilot Token Billing Shift | AI Field Notes #28

    GitHub Copilot moves to AI Credits billing on June 1, ending flat-rate pricing for developers running agent workflows at scale. Google I/O opened today with Gemini 3.1 Flash-Lite at $0.25 per million…
  45. Issue #27 18 May 2026

    Copilot Token Billing | AI Field Notes #27

    GitHub Copilot moves to per-token billing on June 1, replacing request allotments with AI Credits at $0.01 each; heavier models cost more, and the community's early read is that monthly Pro users wil…
  46. Issue #26 16 May 2026

    Agentic Coding Surge | AI Field Notes #26

    AI coding agents crossed into production this week as Q1 2026 tech layoffs reached 80,000, with 47.9% attributed to AI automation. OpenAI shipped Codex on iOS and Android for remote supervision of lo…
  47. Issue #25 15 May 2026

    Open-Weights Cost Shock | AI Field Notes #25

    Four Chinese labs released open-weights coding models in 12 days at under a third of Claude Opus 4.7's price, upending model choice for agent builders. Anthropic published a paper claiming 'evil AI'…
  48. Issue #24 14 May 2026

    AI in everyday software | AI Field Notes #24

    AI in everyday software took a step this week as model availability moved into production tools developers and creators already use. Anthropic shipped Claude across Microsoft 365 (Excel, Word, PowerP…
  49. Issue #23 12 May 2026

    AI Voice API Launch | AI Field Notes #23

    OpenAI launched three new models in its Realtime API on May 8, including GPT-Realtime-2, the first voice model with GPT-5-class reasoning, priced at $32 per million audio input tokens and available i…
  50. Issue #22 11 May 2026

    AI Coding Stack Shift | AI Field Notes #22

    The AI coding stack shift hit five practitioner-level changes this week, from a viral Karpathy-derived CLAUDE.md to managed agent payments and model migrations. Forrest Chang's andrej-karpathy-skills…
  51. Issue #21 10 May 2026

    AI Developer Tool Wave | AI Field Notes #21

    The AI developer tool wave moved five agent integrations into Cursor, Windsurf, Claude Code, and GitHub Copilot between May 8 and 9, while training data and model governance pressure hardened on the…
  52. Issue #20 9 May 2026

    AI Agent SDK Race | AI Field Notes #20

    The AI agent SDK race tightened on May 8 with OpenAI's subagents and Python code mode, NVIDIA's Nemotron 3 Nano Omni open-weights release through Oracle, and Grok 4.3 landing on the same stack a day…
  53. Issue #19 8 May 2026

    Claude Code Limits Doubled | AI Field Notes #19

    Anthropic doubled Claude Code's five-hour rate limits and removed peak-hour throttling for paid subscribers, after a SpaceX compute deal added 300 megawatts (over 220,000 GPUs) at the Colossus 1 data…
  54. Issue #18 7 May 2026

    Wall Street AI Agents Push | AI Field Notes #18

    The Wall Street AI agents push went from slide deck to product this week, with Anthropic releasing ten finance agent templates that visibly hit FactSet and Morningstar share prices, plus a co-built F…
  55. Issue #17 6 May 2026

    Agent Stack Lock-in Wars | AI Field Notes #17

    Enterprise AI vendor lock-in is hardening fast as Anthropic, OpenAI, and their partners deploy prebuilt agent templates directly into finance, compliance, and customer service workflows. The winning…
  56. Issue #16 5 May 2026

    AI Coding Cost Squeeze | AI Field Notes #16

    AI coding cost squeeze is reshaping how developers pick agents this month, with GitHub Copilot moving every plan to token billing on June 1 and annual Opus 4.7 multipliers jumping from 7.5x to 27x. A…
  57. Issue #15 4 May 2026

    AI Stack Reshuffle | AI Field Notes #15

    The AI stack reshuffled this weekend, with OpenAI's GPT-5.5 and Codex going live on AWS Bedrock days before reports surfaced that OpenAI missed its own revenue and user targets and may struggle to fu…
  58. Issue #14 3 May 2026

    AI Agent Autonomy Crunch | AI Field Notes #14

    AI agents are moving into production faster than organizations can secure them. Mistral's cloud coding agents, Hightouch's marketing workflow bots, and Netomi's customer service tools all shipped thi…
  59. Issue #13 2 May 2026

    AI Coding Agent Push | AI Field Notes #13

    AI coding agents took a real step out of the chat window this week. OpenAI's Codex now drives the desktop in the background, opening pull requests and Jira tickets while you work elsewhere, and Anthr…
  60. Issue #12 1 May 2026

    AI Tooling Cost Squeeze | AI Field Notes #12

    AI tooling costs are reshuffling who pays for what, with GitHub Copilot moving every paid plan to token-based billing on June 1 just as Anthropic eyes a $900 billion valuation off its coding business…
  61. Issue #11 30 April 2026

    AI Execution Risk | AI Field Notes #11

    AI execution risk is moving from research to operations. Meta's capex surge turned a revenue beat into a selloff because investors see spending without returns. China froze robotaxi permits after a c…
  62. Issue #10 29 April 2026

    AI Growth Hits Reality | AI Field Notes #10

    OpenAI's missed revenue targets and compute spending commitments sent AI infrastructure stocks tumbling, forcing the market to ask whether the pace of AI capex can actually be supported by current ea…
  63. Issue #9 28 April 2026

    AI Capital Concentration | AI Field Notes #9

    The pattern across the day was simple: power and money are concentrating fast at the top of the AI stack while everything below is being squeezed for cash. Microsoft and OpenAI dissolved their cloud…
  64. Issue #8 27 April 2026

    AI Agent Autonomy Control | AI Field Notes #8

    Copilot can finally edit your files instead of just suggesting changes, autonomous coding agents like Devin are raising hundreds of millions based on unproven safety claims, and the technical standar…
  65. Issue #7 26 April 2026

    AI Infrastructure Bottleneck | AI Field Notes #7

    The AI boom is running into the unsexy infrastructure layer, and that is where the real constraint is forming. Vast Data, a storage software company, just hit a $30 billion valuation because GPUs sit…
  66. Issue #6 25 April 2026

    AI Infrastructure Bottleneck | AI Field Notes #6

    Today's pattern was the bill arriving. Google committed up to $40 billion to Anthropic, OpenAI doubled the price of its newest model, Meta cut 8,000 jobs to fund $115 billion of AI spending, and Deep…
  67. Issue #5 23 April 2026

    AI Infrastructure Arms Race | AI Field Notes #5

    The AI industry is openly admitting what used to stay buried: the real money is in building the infrastructure underneath, not the models on top. Google is splitting its custom chips to undercut Nvid…
  68. Issue #4 22 April 2026

    AI Capability Outpacing Governance | AI Field Notes #4

    The picture getting clearer this week is that AI capability keeps accelerating while the people and systems around it are stuck playing catch-up. OpenAI shipped a better image generator that eats des…
  69. Issue #3 21 April 2026

    AI Cost Reality Check | AI Field Notes #3

    Every company outside the AI labs is discovering that AI costs and capabilities move faster than budgets, hiring plans, and power grids can handle. Google is ditching single-chip vendors to squeeze c…
  70. Issue #2 20 April 2026

    AI Accountability in Practice | AI Field Notes #2

    Today's pattern: the model-race headlines are the least useful story. The real moves are underneath, in courts, state attorneys general, EU regulators, and activist investors, deciding who is account…
  71. Issue #1 19 April 2026

    AI Capex vs. Workforce Contraction | AI Field Notes #1

    Today's news rhymes in an uncomfortable way: the capex keeps rising, the workforce keeps getting cut, and the polite language around both is wearing thin. Meta reframed 8,000 layoffs as an AI strateg…

Want this in your inbox?

One email a day, built for decisions.

A short read every morning: what actually changed in AI, and what it means for strategy, teams, and risk.