https://jamesm.blog/ai/ Recent content in Artificial Intelligence on jamesm.blog https://jamesm.blog/papermod-cover.png https://jamesm.blog/papermod-cover.png Hugo en Mon, 13 Jul 2026 21:22:00 +0100 https://jamesm.blog/ai/mac-studio-local-llm-guide/ Mon, 13 Jul 2026 21:22:00 +0100 https://jamesm.blog/ai/mac-studio-local-llm-guide/ A practical guide to Mac Studio configs for running popular free models locally (Qwen, LLaMA, Mixtral), realistic performance expectations, and which hardware actually makes sense. https://jamesm.blog/ai/apple-openai-musk-altman-lawsuit/ Sun, 12 Jul 2026 22:00:00 +0100 https://jamesm.blog/ai/apple-openai-musk-altman-lawsuit/ Apple sued OpenAI on July 10, 2026 alleging trade secret theft tied to its hardware ambitions. Two days later, Elon Musk and Sam Altman were trading insults on X - the latest round in a feud that predates the lawsuit by years. https://jamesm.blog/ai/mechanistic-interpretability-inside-the-black-box/ Tue, 07 Jul 2026 09:00:00 +0100 https://jamesm.blog/ai/mechanistic-interpretability-inside-the-black-box/ What mechanistic interpretability actually is, why superposition makes a neural network so hard to read, how sparse autoencoders and circuit tracing cracked open the first real windows into a production model, and why - as a hobbyist looking in from outside - I think this is the most important AI research nobody outside the labs is watching closely enough. https://jamesm.blog/ai/redeploying-fable-5/ Wed, 01 Jul 2026 08:00:00 +0100 https://jamesm.blog/ai/redeploying-fable-5/ Eighteen days after a US export control directive pulled Fable 5 and Mythos 5 offline, Anthropic is redeploying Fable 5 globally. The company says the jailbreak that triggered the recall was minor, that rival models could do the same thing, and that the industry needs a CVSS-style framework for AI jailbreaks. https://jamesm.blog/ai/four-futures-series/ Tue, 30 Jun 2026 08:00:00 +0100 https://jamesm.blog/ai/four-futures-series/ A reading path through the Four Futures series - the framework, live signals, and century-scale projections for where AI reshapes work, power, and meaning. https://jamesm.blog/ai/cursor-ios-mobile-app/ Mon, 29 Jun 2026 21:30:00 +0100 https://jamesm.blog/ai/cursor-ios-mobile-app/ Cursor shipped a native iOS app on June 29, 2026, letting developers launch cloud agents, steer their desktop machine, review diffs, and merge PRs from a phone. The interesting part isn't the app - it's what it says about where the work has moved. https://jamesm.blog/ai/five-archetypes-post-role-team/ Mon, 29 Jun 2026 20:30:00 +0100 https://jamesm.blog/ai/five-archetypes-post-role-team/ Boris Cherny, who created Claude Code, sketched five archetypes he sees on his team as engineering, product, design, and data science melt into one role. Here is what that framing gets right, and where I think it still leaves us guessing. https://jamesm.blog/ai/openai-ipo-chatgpt-market-share/ Sat, 27 Jun 2026 07:30:00 +0100 https://jamesm.blog/ai/openai-ipo-chatgpt-market-share/ OpenAI filed its confidential S-1 on June 8, 2026, targeting a $1 trillion public listing. The same week, ChatGPT fell below 50% market share for the first time. The company spends $2.22 for every dollar it earns. Here is what public markets are actually being asked to buy. https://jamesm.blog/ai/spacex-cursor-acquisition/ Tue, 16 Jun 2026 21:30:00 -0700 https://jamesm.blog/ai/spacex-cursor-acquisition/ SpaceX filed a $60 billion all-stock acquisition of Cursor, the AI-powered code editor, signaling a strategic shift in how tech companies build critical infrastructure for AI development. https://jamesm.blog/ai/evaluating-agents-in-production-trajectory-metrics/ Sun, 14 Jun 2026 09:00:00 +0100 https://jamesm.blog/ai/evaluating-agents-in-production-trajectory-metrics/ Endpoint scoring certifies answers, not behaviour. A practitioner guide to step-level agent evaluation - trajectory rubrics, replay harnesses, and regression suites for multi-step production agents. https://jamesm.blog/ai/world-models-after-language/ Sat, 13 Jun 2026 10:00:00 +0100 https://jamesm.blog/ai/world-models-after-language/ Language-only models have a statistical shadow of physical reality, not a reliable simulator of it. A grounded look at world models - what they are, where the 2026 frontier sits, and why they matter for agents that need to act rather than answer. https://jamesm.blog/ai/fable-mythos-government-suspension/ Sat, 13 Jun 2026 07:30:00 +0100 https://jamesm.blog/ai/fable-mythos-government-suspension/ Four days after Fable 5 and Mythos 5 went public, the US government issued an export control directive ordering Anthropic to suspend all access. Anthropic says it's a misunderstanding over a narrow jailbreak. Here's what happened, what's at stake, and why it matters for every frontier lab. https://jamesm.blog/ai/ai-evals-are-broken/ Fri, 12 Jun 2026 19:00:00 +0100 https://jamesm.blog/ai/ai-evals-are-broken/ A grounded look at why the public benchmark numbers you see on AI model launches in 2026 are increasingly disconnected from real-world capability - saturation, contamination, gaming, and what's replacing them as the basis for actual decisions. https://jamesm.blog/ai/expertise-and-work/ Fri, 12 Jun 2026 08:00:00 +0100 https://jamesm.blog/ai/expertise-and-work/ Reading path on expertise after AI - credentials, judgement, taste, engineering roles, and what being expert might mean by 2030. https://jamesm.blog/ai/what-it-means-to-be-expert-in-2030/ Fri, 12 Jun 2026 08:00:00 +0100 https://jamesm.blog/ai/what-it-means-to-be-expert-in-2030/ A sequel to the credential crisis - how the meaning of expertise is changing as AI assistance becomes ambient, with a concrete walkthrough of what a 2030 civil engineer might actually need to be expert at. https://jamesm.blog/ai/anthropic-circuit-bloomberg-965-billion/ Fri, 12 Jun 2026 06:00:00 +0100 https://jamesm.blog/ai/anthropic-circuit-bloomberg-965-billion/ Bloomberg's The Circuit goes inside Anthropic as it eclipses OpenAI to become the world's most valuable AI lab. Emily Chang's rare access with Dario and Daniela Amodei reveals whether safety-first is a strategy or a conviction - and the Pentagon battle is the answer. https://jamesm.blog/ai/securing-ai-agents/ Thu, 11 Jun 2026 10:00:00 +0100 https://jamesm.blog/ai/securing-ai-agents/ The moment an agent holds real tool access, reliability and security stop being separate problems. A practitioner walkthrough of the confused-deputy attack surface, MCP hardening patterns, and the defenses I actually run on my home agent stack. https://jamesm.blog/ai/policy-on-the-ai-exponential/ Thu, 11 Jun 2026 06:23:00 +0100 https://jamesm.blog/ai/policy-on-the-ai-exponential/ Dario Amodei's new essay argues that slow-moving institutions are now the binding constraint on a fast-moving technology, and lays out a five-part policy agenda for the exponential. A summary of the argument, what lands, and where I still have questions. https://jamesm.blog/ai/when-machines-stop-speaking-our-language/ Wed, 10 Jun 2026 20:30:00 +0100 https://jamesm.blog/ai/when-machines-stop-speaking-our-language/ Human language between AI agents is a compatibility layer, and so are programming languages. A speculative look at machines inventing their own protocols, agents reasoning in neuralese, and a future where software is built without source code a human ever reads - and what we lose if the audit trail goes dark. https://jamesm.blog/ai/claude-fable-5-mythos-5/ Tue, 09 Jun 2026 19:00:00 +0100 https://jamesm.blog/ai/claude-fable-5-mythos-5/ Anthropic launches Claude Fable 5 for general use and Claude Mythos 5 for trusted partners - the same underlying model, split by safeguards. State-of-the-art on coding, vision, and science, with classifiers that fall back to Opus 4.8 on risky topics. https://jamesm.blog/ai/what-im-researching-in-ai-right-now/ Mon, 08 Jun 2026 23:00:00 +0100 https://jamesm.blog/ai/what-im-researching-in-ai-right-now/ I treat my own learning like a research agenda - a small set of questions I am actively chasing rather than a reading list. Here are the AI areas I have been deep in, the ones I have decided to move into next, and the ones I am deliberately leaving on the watch list. https://jamesm.blog/ai/geoffrey-hinton-interviews/ Mon, 08 Jun 2026 21:07:21 +0100 https://jamesm.blog/ai/geoffrey-hinton-interviews/ A working index of Geoffrey Hinton interviews - the Nobel and Turing Award winner often called the Godfather of AI - tracing his arc from backpropagation and deep learning to public warnings about superintelligence, job displacement, and existential risk. https://jamesm.blog/ai/trust-series/ Mon, 08 Jun 2026 08:00:00 +0100 https://jamesm.blog/ai/trust-series/ A reading path through the Trust series - research map, broken benchmarks, agent security, world models, and trajectory evaluation - for builders who need to hand real work to non-deterministic systems. https://jamesm.blog/ai/recursive-self-improvement/ Thu, 04 Jun 2026 20:03:00 +0100 https://jamesm.blog/ai/recursive-self-improvement/ Recursive self-improvement is the idea that an AI could rewrite itself into something smarter, then repeat - the engine behind every intelligence-explosion scenario. Here is what the theory actually claims, which pieces are already real in 2026, the bottlenecks that decide whether it runs away or fizzles, and why the honest answer is partly, slowly, and not the way the science fiction tells it. https://jamesm.blog/ai/the-computational-case-for-consciousness/ Sun, 31 May 2026 10:30:00 +0100 https://jamesm.blog/ai/the-computational-case-for-consciousness/ I wrote up the case that consciousness is fundamental through Donald Hoffman. Honesty demands the other side: the case that consciousness is computational - something that switches on when information is processed in the right way. Global Workspace Theory, Integrated Information Theory, the hard problem, and why I still cannot close the question. https://jamesm.blog/ai/donald-hoffman-consciousness-fundamental/ Fri, 29 May 2026 10:30:00 +0100 https://jamesm.blog/ai/donald-hoffman-consciousness-fundamental/ Donald Hoffman argues that consciousness is fundamental and the physical world is an evolved interface - and he tries to prove it with evolutionary game theory and a formal mathematics of conscious agents. A walk through the argument, the maths, the objections, and why I keep coming back to it. https://jamesm.blog/ai/yampolskiy-personal-universes/ Fri, 29 May 2026 08:30:00 +0100 https://jamesm.blog/ai/yampolskiy-personal-universes/ Roman Yampolskiy's Personal Universes proposal takes the simulation hypothesis seriously and turns it into an alignment strategy: one optimised reality per person. A look at what the idea claims, where it cracks, and why it is more useful than it looks. https://jamesm.blog/ai/will-ai-kill-coding-jobs-cherny/ Tue, 26 May 2026 08:00:00 +0100 https://jamesm.blog/ai/will-ai-kill-coding-jobs-cherny/ Boris Cherny, creator of Claude Code, sits down with Sky News and reacts to three charts about AI, jobs, and the future of software. A look at what is genuinely new in his argument, what is well-worn, and where the printing press analogy actually holds up. https://jamesm.blog/ai/ai-cyber-threat-is-rising/ Tue, 26 May 2026 07:30:00 +0100 https://jamesm.blog/ai/ai-cyber-threat-is-rising/ Sky News technology correspondent Rowland Manthorpe lays out the evidence that AI models are getting dramatically better at offensive cyber tasks. A look at what the AISI and CAISI numbers actually say, why open-weight releases change the threat curve, and what defenders should do about it. https://jamesm.blog/ai/context-engineering/ Wed, 20 May 2026 20:45:00 +0100 https://jamesm.blog/ai/context-engineering/ Prompt engineering was about wording a single request well. Context engineering is about curating everything in the model's window - retrieval, memory, tool results, and compaction - across a whole agent run. Here is why the discipline shifted, and the practices that separate reliable production systems from ones that quietly rot. https://jamesm.blog/ai/ai-dev-tooling/ Wed, 20 May 2026 08:00:00 +0100 https://jamesm.blog/ai/ai-dev-tooling/ Canonical reading path for AI-assisted development - stack decisions, learning order, spec-driven workflows, and deep dives on Cursor vs Claude Code. https://jamesm.blog/ai/ai-economics-hardware/ Mon, 18 May 2026 08:00:00 +0100 https://jamesm.blog/ai/ai-economics-hardware/ Reading path for intelligence economics - token costs, local vs cloud, GPU choices, energy constraints, and the inference hardware insurgents. https://jamesm.blog/ai/cursor-composer-2-5/ Mon, 18 May 2026 07:30:00 +0100 https://jamesm.blog/ai/cursor-composer-2-5/ Cursor's new Composer 2.5 is the first time the company's in-house coding model deserves to be treated as a daily driver rather than a budget fallback. A look at what changed, what it costs, and when to reach for it instead of Claude, GPT-5.5, or Gemini. https://jamesm.blog/ai/ai-as-analogy-engine/ Sat, 16 May 2026 08:02:00 +0000 https://jamesm.blog/ai/ai-as-analogy-engine/ The deepest misunderstanding about modern AI is that it merely retrieves memorised text. The more interesting truth is that large models form analogies, combine concepts, and explore conceptual relationships at a scale that may eventually industrialise parts of the invention process itself. https://jamesm.blog/ai/home-agent-stack/ Fri, 15 May 2026 08:00:00 +0100 https://jamesm.blog/ai/home-agent-stack/ A reading path for building a home AI agent on Apple Silicon - local inference, MCP servers, persistent memory, remote access, and security hardening in order. https://jamesm.blog/ai/agent-reliability-debugging-non-deterministic/ Fri, 15 May 2026 06:30:00 +0100 https://jamesm.blog/ai/agent-reliability-debugging-non-deterministic/ AI agents that work in the demo and fail in production are the standard story of the last two years. A practical look at why agent reliability is genuinely hard, what patterns have emerged for handling it, and which testing approaches actually work. https://jamesm.blog/ai/dario-amodei-anthropic-ceo/ Thu, 14 May 2026 21:22:00 +0100 https://jamesm.blog/ai/dario-amodei-anthropic-ceo/ A grounded look at Dario Amodei - the Princeton-trained physicist who left OpenAI to co-found Anthropic, made AI safety the company's commercial differentiator, and now runs the second-most-valuable private AI company on the planet. https://jamesm.blog/ai/ai-in-scientific-research-2026/ Wed, 13 May 2026 06:00:00 +0100 https://jamesm.blog/ai/ai-in-scientific-research-2026/ AlphaFold was the headline AI-for-science story. In 2026 the story has broadened into something more interesting - AI tools in routine use across many scientific fields, with uneven but real impact. A practical look at where the gains actually are. https://jamesm.blog/ai/state-of-open-weight-models-2026/ Tue, 12 May 2026 09:00:00 +0100 https://jamesm.blog/ai/state-of-open-weight-models-2026/ A grounded look at where the open-weight LLM ecosystem actually stands in May 2026 - Llama 4, Qwen 3.5, DeepSeek V4, Gemma 4, Mistral Medium 3.5 - what each family is best at, where the licences land, and how close the gap with the frontier proprietary models has actually closed. https://jamesm.blog/ai/ai-energy-crisis-data-center-power/ Mon, 11 May 2026 17:00:00 +0100 https://jamesm.blog/ai/ai-energy-crisis-data-center-power/ A grounded look at the energy constraint becoming the binding limit on AI deployment in 2026 - the power demand, the transformer backlogs, the grid interconnect queues, the rising consumer electricity prices, and what it means for the shape of the next decade of compute. https://jamesm.blog/ai/inference-hardware-insurgents/ Mon, 11 May 2026 13:00:00 +0100 https://jamesm.blog/ai/inference-hardware-insurgents/ A grounded look at the three serious AI inference hardware insurgents in 2026 - Cerebras with its wafer-scale WSE-3, Groq with its LPU architecture now part-licensed by Nvidia, and SambaNova with its SN40L and SN50 reconfigurable dataflow units - and what their different bets tell us about where inference is heading. https://jamesm.blog/ai/multimodal-ai-2026/ Sat, 09 May 2026 08:00:00 +0100 https://jamesm.blog/ai/multimodal-ai-2026/ Where multimodal AI has actually delivered value in 2026, where it is still demoware, and the patterns that determine whether vision and audio are quietly indispensable in your stack or just an expensive distraction. https://jamesm.blog/ai/prompt-caching/ Sat, 09 May 2026 08:00:00 +0100 https://jamesm.blog/ai/prompt-caching/ What prompt caching actually is, why it is the highest-leverage performance and cost optimisation available to most LLM applications, and the patterns that determine whether you get a 90 percent cost reduction or no benefit at all. https://jamesm.blog/ai/reasoning-models-2026/ Fri, 08 May 2026 19:00:00 +0100 https://jamesm.blog/ai/reasoning-models-2026/ A grounded look at the reasoning-model wave - what makes o3, DeepSeek R1/R2, Gemini Deep Think, and Claude Extended Thinking different from the generation that came before, when to actually reach for one, and what the cost and capability trade-offs look like in production. https://jamesm.blog/ai/scott-galloway-on-ai/ Mon, 04 May 2026 22:31:00 +0100 https://jamesm.blog/ai/scott-galloway-on-ai/ A grounded look at Scott Galloway - the NYU marketing professor and serial entrepreneur arguing that AI is a late-stage bubble, that AI was built for the rich, and that storytelling and relationships are the skills the next decade rewards. https://jamesm.blog/ai/agent-protocols-mcp-a2a-acp/ Sun, 03 May 2026 12:00:00 +0100 https://jamesm.blog/ai/agent-protocols-mcp-a2a-acp/ While the headlines were about model launches, the agent ecosystem quietly converged on three protocols that matter - MCP, A2A, and ACP. A grounded comparison of what each one actually does, where they overlap, and which to bet on if you are building agents in 2026. https://jamesm.blog/ai/llm-powered-personal-productivity/ Sun, 03 May 2026 08:00:00 +0100 https://jamesm.blog/ai/llm-powered-personal-productivity/ Local LLMs have quietly become good enough to run a real personal automation stack - inbox triage, second-brain notes, home routines, voice capture - without sending anything to the cloud. A walkthrough of what I run, why each piece is there, and where the seams are. https://jamesm.blog/ai/roman-yampolskiy/ Sat, 02 May 2026 21:03:00 +0100 https://jamesm.blog/ai/roman-yampolskiy/ A grounded look at Roman Yampolskiy - the computer scientist who coined the term AI safety, why he argues superintelligence is fundamentally uncontrollable, and how seriously to take him. https://jamesm.blog/ai/humanoid-robotics-2026/ Sat, 02 May 2026 09:00:00 +0100 https://jamesm.blog/ai/humanoid-robotics-2026/ After a decade of prototypes, humanoid robots are crossing into early production in 2026. A technical look at the leading platforms, the economics, the actuator breakthroughs, and what is still hype. https://jamesm.blog/ai/ai-agents-that-actually-work/ Fri, 01 May 2026 08:00:00 +0100 https://jamesm.blog/ai/ai-agents-that-actually-work/ The patterns that separate agents that ship and stay shipped from the ones that demo well and fall over the moment they meet a real workload. https://jamesm.blog/ai/ai-tools/ Fri, 01 May 2026 06:00:00 +0100 https://jamesm.blog/ai/ai-tools/ A categorised, curated index of AI tools across art, chatbots, coding, music, video, voice, writing, research, and local LLMs - updated for 2026. https://jamesm.blog/ai/ai-safety-first-principles/ Thu, 30 Apr 2026 08:00:00 +0100 https://jamesm.blog/ai/ai-safety-first-principles/ Stripping the AI safety conversation down to its load-bearing parts, separating the engineering problems that matter today from the speculative scenarios that absorb most of the attention. https://jamesm.blog/ai/ai-skills-one-folder-any-model/ Thu, 30 Apr 2026 06:32:00 +0100 https://jamesm.blog/ai/ai-skills-one-folder-any-model/ What Claude Code skills actually are under the hood, why the SKILL.md format is model-agnostic, and how the same folder can drive Claude, Gemini, Codex, and a dozen other agents without changes. https://jamesm.blog/ai/ai-augmented-design-workflow/ Wed, 29 Apr 2026 08:30:00 +0100 https://jamesm.blog/ai/ai-augmented-design-workflow/ How a combination of Cursor in the IDE, Claude Code and Codex in the terminal, and ChatGPT for general questions, collapsed my discuss-design-document loop to under ten minutes - backed by transcribed meetings checked into GitHub and a daily cross-check between code and design. https://jamesm.blog/ai/fine-tune-vs-rag/ Wed, 29 Apr 2026 08:00:00 +0100 https://jamesm.blog/ai/fine-tune-vs-rag/ A practical decision framework for choosing between fine-tuning and retrieval-augmented generation, the failure modes of picking the wrong one, and the hybrid patterns that quietly do most of the heavy lifting in production. https://jamesm.blog/ai/free-intelligence-era/ Tue, 28 Apr 2026 00:48:00 +0100 https://jamesm.blog/ai/free-intelligence-era/ The single most important economic fact about AI is that the marginal cost of intelligence is collapsing toward zero. Most of our institutions, careers, and pricing models were built for the regime where it was scarce. This is what breaks when thinking goes from expensive to nearly free, and what that future actually rewards. https://jamesm.blog/ai/ai-hallucinations-understanding-and-mitigating/ Tue, 28 Apr 2026 00:02:00 +0100 https://jamesm.blog/ai/ai-hallucinations-understanding-and-mitigating/ What hallucinations actually are, why the word is a slightly misleading import from psychiatry, why no model will ever stop producing them entirely, and the small set of techniques that genuinely move the dial on false outputs in production. https://jamesm.blog/ai/phone-your-home-ai-agent/ Mon, 27 Apr 2026 21:48:00 +0100 https://jamesm.blog/ai/phone-your-home-ai-agent/ A practical walkthrough of the stack I use to literally phone my Mac Studio, speak to a local AI agent, and get it to run or check work while I'm away from the desk. https://jamesm.blog/ai/mcp-servers-home-ai-agent/ Mon, 27 Apr 2026 00:12:00 +0100 https://jamesm.blog/ai/mcp-servers-home-ai-agent/ The voice pipeline and the local model are the easy part. The actual power comes from the tools the agent can reach. A walkthrough of the MCP servers I run on my Mac Studio to give my home agent filesystem, email, calendar, notes, and home-automation access - and the choices that stopped it from becoming a security liability. https://jamesm.blog/ai/the-year-3026/ Sun, 26 Apr 2026 22:03:00 +0100 https://jamesm.blog/ai/the-year-3026/ A thousand years is long enough that nothing we currently call a company, a currency, or a country is likely to survive. What does thinking seriously about 3026 actually look like - and what, if anything, holds its value across that horizon? https://jamesm.blog/ai/the-year-2126/ Sun, 26 Apr 2026 21:02:00 +0100 https://jamesm.blog/ai/the-year-2126/ A hundred years is close enough to reason about concretely and far enough that most of what feels fixed today will be gone. What does 2126 honestly look like - and which of today's problems have we solved by then, which have we inherited, and which are entirely new? https://jamesm.blog/ai/reading-the-signals-four-futures/ Sat, 25 Apr 2026 20:45:00 +0100 https://jamesm.blog/ai/reading-the-signals-four-futures/ A follow-up to the four futures framework - scoring the real-world signals firing right now against broad abundance, winner-take-most, techno-feudalism, and managed transition. https://jamesm.blog/ai/the-next-decade-of-ai/ Sat, 25 Apr 2026 07:55:00 +0100 https://jamesm.blog/ai/the-next-decade-of-ai/ Most predictions about the future of AI are either utopian or apocalyptic. The real trajectory is stranger and more interesting than both - and it's already visible if you look at what's quietly working right now. https://jamesm.blog/ai/grok-voice-apis/ Fri, 24 Apr 2026 21:07:00 +0000 https://jamesm.blog/ai/grok-voice-apis/ xAI launches standalone Speech-to-Text and Text-to-Speech APIs, bringing production-grade voice capabilities to developers at competitive pricing. https://jamesm.blog/ai/gpt-5-5-release/ Fri, 24 Apr 2026 07:51:00 +0100 https://jamesm.blog/ai/gpt-5-5-release/ OpenAI shipped GPT-5.5 on April 23, 2026 - the first fully retrained base model since GPT-4.5. A look at the benchmarks, the new features, and whether this release actually matters. https://jamesm.blog/ai/agent-first-architecture-engineer-as-curator/ Thu, 23 Apr 2026 21:01:00 +0100 https://jamesm.blog/ai/agent-first-architecture-engineer-as-curator/ Some thoughts on where agent-first architecture might take us, and what engineering could look like if the job shifts from building systems line by line to curating, governing, and evolving fleets of agents. https://jamesm.blog/ai/chatgpt-images-2/ Thu, 23 Apr 2026 06:52:00 +0100 https://jamesm.blog/ai/chatgpt-images-2/ ChatGPT Images 2.0 arrives with thinking mode, web-search-informed generation, and text rendering that actually works. Here's what changed and why the reaction has been so positive. https://jamesm.blog/ai/ai-law-trends-2026/ Wed, 22 Apr 2026 21:00:00 +0100 https://jamesm.blog/ai/ai-law-trends-2026/ The legal frameworks around AI are shifting fast. The EU AI Act is in force, copyright battles are mounting in US courts, and liability questions are landing on businesses whether they are ready or not. https://jamesm.blog/ai/home-ai-agent-memory-that-lasts/ Wed, 22 Apr 2026 19:30:00 +0100 https://jamesm.blog/ai/home-ai-agent-memory-that-lasts/ An agent with tools but no memory forgets you every morning. A walkthrough of how I bolted durable, structured memory onto my home AI agent running on a Mac Studio - what I store, where it lives, how retrieval works, and the mistakes I stopped making after the third rewrite. https://jamesm.blog/ai/ai-tooling-learning-path/ Tue, 21 Apr 2026 19:07:00 +0100 https://jamesm.blog/ai/ai-tooling-learning-path/ A phased learning path for getting from zero to a full AI development stack in 2026, without drowning in tool choice or skipping the fundamentals. https://jamesm.blog/ai/amazon-anthropic-25-billion-investment/ Tue, 21 Apr 2026 06:36:00 +0100 https://jamesm.blog/ai/amazon-anthropic-25-billion-investment/ Amazon announces up to $25 billion additional investment in Anthropic, with Anthropic committing over $100 billion to AWS over the next decade. https://jamesm.blog/ai/hermes-agent/ Mon, 20 Apr 2026 19:03:00 +0100 https://jamesm.blog/ai/hermes-agent/ Nous Research's Hermes Agent is an open-source, persistent autonomous system. A look at features, use cases, pricing, and how it compares to OpenClaw, Claude Managed Agents, and Perplexity Computer. https://jamesm.blog/ai/mac-dictation-tools-comparison/ Mon, 20 Apr 2026 19:02:00 +0100 https://jamesm.blog/ai/mac-dictation-tools-comparison/ A practical breakdown of the three dictation apps every Mac user keeps hearing about - who they're for, what they cost, and where they stop being interchangeable. https://jamesm.blog/ai/ai-cloud-subscriptions/ Sun, 19 Apr 2026 07:12:00 +0100 https://jamesm.blog/ai/ai-cloud-subscriptions/ A 2026 buyer's guide to AI cloud subscriptions. Compare API pricing, context windows, rate limits, and consumer plans across Anthropic, OpenAI, Google, Microsoft, xAI, Mistral, and DeepSeek. https://jamesm.blog/ai/dgx-spark-vs-mac-studio/ Sun, 19 Apr 2026 05:22:00 +0000 https://jamesm.blog/ai/dgx-spark-vs-mac-studio/ An in-depth comparison of NVIDIA DGX Spark and Apple Mac Studio for local LLM inference. Includes specifications, pricing, performance benchmarks, and cost analysis to help you choose the right personal AI supercomputer. https://jamesm.blog/ai/ai-resources-and-best-practices/ Sat, 18 Apr 2026 08:15:00 +0100 https://jamesm.blog/ai/ai-resources-and-best-practices/ A comprehensive guide to the best resources and practices for working effectively with AI models, from prompt engineering to system design. https://jamesm.blog/ai/claude-token-efficiency-mindset/ Fri, 17 Apr 2026 21:36:00 +0000 https://jamesm.blog/ai/claude-token-efficiency-mindset/ Most people waste tokens on Claude not because they're bad at prompting, but because they haven't internalized the real cost structure. Here's how to think about token efficiency as a first-class design constraint. https://jamesm.blog/ai/claude-design/ Fri, 17 Apr 2026 19:11:00 +0000 https://jamesm.blog/ai/claude-design/ Claude Design brings AI-powered visual collaboration to teams, automating the tedious handoff between design and development while maintaining design system consistency. https://jamesm.blog/ai/four-futures-machine-speed-economy/ Thu, 16 Apr 2026 19:26:00 +0100 https://jamesm.blog/ai/four-futures-machine-speed-economy/ A map of four plausible futures - broad abundance, winner-take-most, techno-feudalism, and a managed transition - with the signals that suggest which one is emerging. https://jamesm.blog/ai/claude-opus-4-7/ Thu, 16 Apr 2026 16:15:00 +0000 https://jamesm.blog/ai/claude-opus-4-7/ Claude Opus 4.7 brings autonomous software engineering, 3.75 megapixel vision, and extended reasoning control - raising the bar for AI-assisted development. https://jamesm.blog/ai/open-webui-self-hosted-llm-interface/ Wed, 15 Apr 2026 23:15:00 +0000 https://jamesm.blog/ai/open-webui-self-hosted-llm-interface/ Open WebUI is an open-source interface that brings ChatGPT-like convenience to local language models, giving you a sleek chat interface for Ollama, OpenAI, and compatible APIs with zero vendor lock-in. https://jamesm.blog/ai/paperless-ngx-self-hosted-document-management/ Wed, 15 Apr 2026 22:21:00 +0000 https://jamesm.blog/ai/paperless-ngx-self-hosted-document-management/ Paperless-ngx is a self-hosted document management system that lets you scan, digitize, and organize physical paperwork with full control over your data and zero subscription fees. https://jamesm.blog/ai/ollama/ Tue, 14 Apr 2026 08:25:25 +0100 https://jamesm.blog/ai/ollama/ A practical guide to Ollama, the platform for running large language models locally, including how to launch OpenClaw and explore its available models. https://jamesm.blog/ai/github-backing-openclaw/ Tue, 14 Apr 2026 08:05:00 +0100 https://jamesm.blog/ai/github-backing-openclaw/ GitHub became an official sponsor of OpenClaw, the fastest-growing open source project in history, providing Copilot Pro+ and security funding. https://jamesm.blog/ai/token-economics-why-costs-arent-going-down/ Mon, 13 Apr 2026 06:40:00 +0100 https://jamesm.blog/ai/token-economics-why-costs-arent-going-down/ Despite massive scale and competition, LLM pricing hasn't meaningfully declined. Here's why the fundamentals don't support cheaper AI. https://jamesm.blog/ai/claude-mythos-restricted/ Mon, 13 Apr 2026 06:12:00 +0100 https://jamesm.blog/ai/claude-mythos-restricted/ Anthropic's Claude Mythos shattered performance records but was immediately restricted to a select few, marking a new era where AI's immense power demands unprecedented control. https://jamesm.blog/ai/structured-outputs-schema/ Sun, 12 Apr 2026 06:17:00 +0000 https://jamesm.blog/ai/structured-outputs-schema/ Why guaranteeing output format matters more than you think, and when to use structured outputs in production https://jamesm.blog/ai/small-language-models/ Sun, 12 Apr 2026 06:03:00 +0100 https://jamesm.blog/ai/small-language-models/ Why smaller language models are proving that efficiency, speed, and cost matter more than raw parameter count. A look at the shift redefining AI development. https://jamesm.blog/ai/llm-context-window-arms-race/ Sat, 11 Apr 2026 06:14:00 +0000 https://jamesm.blog/ai/llm-context-window-arms-race/ The obsession with massive context windows misses what actually matters in production LLM systems https://jamesm.blog/ai/local-vs-cloud-ai-2026/ Sat, 11 Apr 2026 06:11:00 +0100 https://jamesm.blog/ai/local-vs-cloud-ai-2026/ In 2026, the choice between running AI locally or in the cloud is no longer a binary one. It is a strategic architectural decision based on privacy, latency, and 'reasoning density'. https://jamesm.blog/ai/cline/ Fri, 10 Apr 2026 10:00:00 +0100 https://jamesm.blog/ai/cline/ An exploration of Cline, the autonomous AI coding agent that lives in your IDE and handles complex, multi-step engineering tasks through tool-use and agency. https://jamesm.blog/ai/cline-kanban/ Fri, 10 Apr 2026 07:44:00 +0100 https://jamesm.blog/ai/cline-kanban/ How to integrate Cline with Kanban boards via Model Context Protocol to create a seamless autonomous workflow that bridges development and task management. https://jamesm.blog/ai/ai-reliability-is-weird/ Thu, 09 Apr 2026 22:00:00 +0000 https://jamesm.blog/ai/ai-reliability-is-weird/ Traditional testing methods break down in the age of autonomous AI agents. A practical look at why deterministic testing fails for non-deterministic builders, and what actually works instead. https://jamesm.blog/ai/career-ops/ Thu, 09 Apr 2026 07:25:00 +0000 https://jamesm.blog/ai/career-ops/ Career-Ops inverts the typical job-search power dynamic by giving candidates AI-powered tools to intelligently evaluate and prioritize opportunities, automate application workflows, and maintain strategic control over their career decisions. https://jamesm.blog/ai/claude-mythos-benchmarks/ Wed, 08 Apr 2026 08:45:00 +0100 https://jamesm.blog/ai/claude-mythos-benchmarks/ Claude Mythos Preview shattered AI benchmarks with 93.9% SWE-bench and 97.6% USAMO scores, but remains restricted to security partners due to cybersecurity risks. https://jamesm.blog/ai/claude-code-vs-cursor/ Wed, 08 Apr 2026 07:30:00 +0100 https://jamesm.blog/ai/claude-code-vs-cursor/ After six months of daily use, here is how the two heavyweights of AI-assisted coding compare: the terminal-native Claude Code and the IDE-integrated Cursor. https://jamesm.blog/ai/automation-paradox/ Tue, 07 Apr 2026 14:00:00 +0000 https://jamesm.blog/ai/automation-paradox/ As AI handles routine cognitive work, the irreplaceable skill becomes not execution, but deciding what's worth doing - and why human judgment becomes more valuable, not less. https://jamesm.blog/ai/spec-driven-development/ Tue, 07 Apr 2026 09:00:00 +0000 https://jamesm.blog/ai/spec-driven-development/ With AI handling implementation, the specification itself becomes the product. What you're really selling is clarity of thought - and the ability to write briefs that don't lie. https://jamesm.blog/ai/architect-vs-builder/ Mon, 06 Apr 2026 14:00:00 +0000 https://jamesm.blog/ai/architect-vs-builder/ The traditional engineering hierarchy of junior→senior→architect is collapsing. What emerges is a cleaner, harsher split: people who decide what to build, and people who turn that decision into code. https://jamesm.blog/ai/expertise-after-ai/ Mon, 06 Apr 2026 09:00:00 +0000 https://jamesm.blog/ai/expertise-after-ai/ If AI can pass the bar exam, the CPA exam, and every technical certification, what are we actually certifying? Expertise becomes something else entirely - and much harder to fake. https://jamesm.blog/ai/gpu-servers-vs-api-credits/ Sun, 05 Apr 2026 23:16:25 +0100 https://jamesm.blog/ai/gpu-servers-vs-api-credits/ A practical 2026 breakdown of when to rent GPUs versus buy API credits, with cost-per-token figures, break-even thresholds, and a hybrid architecture playbook. https://jamesm.blog/ai/what-actually-belongs-in-my-ai-dev-stack-2026/ Sun, 05 Apr 2026 00:05:00 +0100 https://jamesm.blog/ai/what-actually-belongs-in-my-ai-dev-stack-2026/ A proper AI dev stack in 2026 needs distinct layers - spec writing, fast editing, heavy agentic work, cheap models, review, research, and capture. The case for treating AI as workflow architecture, not a bolt-on assistant. https://jamesm.blog/ai/github-spec-kit-2026-update/ Sat, 04 Apr 2026 21:06:14 +0100 https://jamesm.blog/ai/github-spec-kit-2026-update/ GitHub Spec Kit v0.5.0 evolved from a documentation toolkit into a full AI development platform in 2026 - Claude Code as a native skill, AWS Kiro integration, and multi-agent support across Cursor, Copilot, and Gemini CLI. https://jamesm.blog/ai/taste-is-the-new-scarcity/ Sat, 04 Apr 2026 12:00:00 +0100 https://jamesm.blog/ai/taste-is-the-new-scarcity/ When AI can generate thousands of solutions on demand, the bottleneck shifts from thinking capacity to judgment. Taste - knowing which answer is actually right - becomes the primary human skill in an age of abundant intelligence. https://jamesm.blog/ai/personal-ai-dev-stack/ Fri, 03 Apr 2026 23:09:25 +0100 https://jamesm.blog/ai/personal-ai-dev-stack/ A practical, AI-driven development stack combining cloud and local LLMs, terminal tools, IDEs, and mobile access for persistent, productive workflows. https://jamesm.blog/ai/twenty-dollar-ai-era-is-over/ Fri, 03 Apr 2026 08:45:00 +0100 https://jamesm.blog/ai/twenty-dollar-ai-era-is-over/ Every major AI provider is quietly repositioning the $20/month tier. Here's what's happening at Anthropic, OpenAI, Google, Microsoft, xAI, and what monthly prices are likely to look like by the end of 2026. https://jamesm.blog/ai/meaning-of-work-age-of-abundance/ Thu, 02 Apr 2026 23:30:00 +0100 https://jamesm.blog/ai/meaning-of-work-age-of-abundance/ If agents end up doing the execution, what are we for? Some thinking out loud about purpose, identity, and what a working life might look like when the labour layer is automated. https://jamesm.blog/ai/spacex-cursor-60-billion-option/ Thu, 02 Apr 2026 19:16:00 +0100 https://jamesm.blog/ai/spacex-cursor-60-billion-option/ SpaceX has taken an option to acquire Cursor for $60 billion later this year, or pay $10 billion for a joint build. A look at the deal structure, what it tells us about xAI's coding ambitions, and why Anysphere is suddenly the most sought-after company in developer tools. https://jamesm.blog/ai/claude-code-source-leak/ Wed, 01 Apr 2026 07:42:00 +0000 https://jamesm.blog/ai/claude-code-source-leak/ Nearly 2,000 internal files from Claude Code were briefly leaked due to human error. Here's what happened, what was exposed, and the security implications for developers using Anthropic's AI coding tool. https://jamesm.blog/ai/meta-employee-tracking/ Wed, 01 Apr 2026 08:30:00 +0100 https://jamesm.blog/ai/meta-employee-tracking/ Meta's Model Capability Initiative is installing monitoring software on US employee laptops to capture keystrokes, clicks, and screenshots - all to train AI agents. There is no opt-out. https://jamesm.blog/ai/year-of-agents-and-whats-next/ Fri, 13 Mar 2026 08:06:00 +0100 https://jamesm.blog/ai/year-of-agents-and-whats-next/ A look back at how AI tooling and agents shifted between April 2025 and April 2026, what is clearly coming in the next two years, and how the changes will reshape both work and personal workflows. https://jamesm.blog/ai/where-should-documentation-live/ Thu, 12 Mar 2026 07:30:00 +0100 https://jamesm.blog/ai/where-should-documentation-live/ A brainstorm - not a proposal - on where different types of documents and artefacts should live, what the source of truth should be, and how AI tooling is reshaping that question. https://jamesm.blog/ai/junior-developer-pipeline-problem/ Thu, 12 Mar 2026 00:12:00 +0100 https://jamesm.blog/ai/junior-developer-pipeline-problem/ Most of the AI-and-developers conversation assumes a steady supply of senior engineers who can curate, review, and judge. But the work that used to produce those seniors is the work AI now does. Where does the next generation come from? https://jamesm.blog/ai/we-are-learning-to-buy-intelligence/ Wed, 11 Mar 2026 07:09:25 +0100 https://jamesm.blog/ai/we-are-learning-to-buy-intelligence/ Intelligence used to be something you hired, trained for, or waited in line for. Now it is becoming infrastructure - metered, on-demand, and collapsing in price. https://jamesm.blog/ai/openclaw-is-wild/ Tue, 10 Mar 2026 07:09:25 +0100 https://jamesm.blog/ai/openclaw-is-wild/ OpenClaw is an open-source AI agent framework that lets language models operate software directly through computer interfaces - clicking, typing, and navigating without custom APIs, making any software automatable. https://jamesm.blog/ai/claude-code-review/ Mon, 09 Mar 2026 07:14:25 +0100 https://jamesm.blog/ai/claude-code-review/ Claude Code's multi-agent code review dispatches parallel AI agents to cross-check PRs from different angles, filtering false positives and ranking issues by severity - a fundamentally different approach to automated review. https://jamesm.blog/ai/claude_code_limits-hybrid_ai_stack/ Mon, 09 Mar 2026 07:09:25 +0100 https://jamesm.blog/ai/claude_code_limits-hybrid_ai_stack/ Hitting Claude Code usage limits doesn't mean upgrading to $200/month. A hybrid AI stack using local models, cheap cloud inference, and frontier models strategically is a smarter and cheaper approach. https://jamesm.blog/ai/github-spec-kit/ Wed, 03 Dec 2025 13:21:14 +0100 https://jamesm.blog/ai/github-spec-kit/ GitHub Spec Kit is a version-controlled markdown framework for spec-driven development - writing the specification first and letting AI tools implement against it to prevent architectural drift. https://jamesm.blog/ai/cursor-ai/ Wed, 03 Dec 2025 13:15:25 +0100 https://jamesm.blog/ai/cursor-ai/ Cursor AI reads your repo with architectural awareness and turns complex refactors into conversations. Combined with GitHub Spec Kit, it creates a tight spec-to-implementation loop. https://jamesm.blog/ai/human-in-demand-jobs/ Thu, 27 Nov 2025 07:39:14 +0100 https://jamesm.blog/ai/human-in-demand-jobs/ Which jobs survive when AI handles execution? A look at the five categories of work that stay human-shaped - judgment, empathy, embodied skill, taste, and accountability. https://jamesm.blog/ai/sergey-brin-interviews/ Tue, 20 May 2025 19:27:14 +0100 https://jamesm.blog/ai/sergey-brin-interviews/ A working index of Sergey Brin interviews on AI, Gemini, robotics, AGI timelines, and the future of human-computer interaction - tracking the Google co-founder's public commentary since his return to hands-on work in 2024-2026. https://jamesm.blog/ai/ai-agents-emergency-debate/ Mon, 12 May 2025 16:27:14 +0100 https://jamesm.blog/ai/ai-agents-emergency-debate/ An emergency debate on AI's impact on employment - which sectors face near-term disruption, which roles stay human-shaped, and how fast the transition is actually moving. https://jamesm.blog/ai/deepseek/ Mon, 27 Jan 2025 21:39:14 +0100 https://jamesm.blog/ai/deepseek/ DeepSeek R1's January 2025 release shook the AI industry - a frontier-grade reasoning model built for a reported $6M that forced a rethink of training costs, open-source strategy, and closed-model pricing. https://jamesm.blog/ai/unitree-g1/ Tue, 21 May 2024 20:09:25 +0100 https://jamesm.blog/ai/unitree-g1/ Unitree's G1 humanoid robot starts at $16k - dramatically undercutting Tesla Optimus and Boston Dynamics while using reinforcement and imitation learning for precision real-world tasks. https://jamesm.blog/ai/chatbots-llms/ Fri, 17 May 2024 17:06:25 +0100 https://jamesm.blog/ai/chatbots-llms/ A practical explainer on the difference between chatbots and large language models, how modern model families fit together, and how to choose the right AI tool for a task. https://jamesm.blog/ai/1x-eve/ Sun, 21 Apr 2024 22:26:14 +0100 https://jamesm.blog/ai/1x-eve/ 1X EVE is a wheeled android worker robot built for logistics and retail environments, using embodied AI and natural language processing to navigate real-world spaces safely alongside humans. https://jamesm.blog/ai/tesla-optimus-robot/ Sun, 21 Apr 2024 20:44:14 +0100 https://jamesm.blog/ai/tesla-optimus-robot/ Tesla Optimus Gen 3 is a bipedal humanoid robot powered by Tesla's vision and self-driving AI stack, targeting under $25,000 per unit - with Elon Musk predicting the robot business will eventually exceed the car business. https://jamesm.blog/ai/boston-dynamics-humanoid-robot/ Sat, 20 Apr 2024 22:09:25 +0100 https://jamesm.blog/ai/boston-dynamics-humanoid-robot/ Boston Dynamics' all-electric Atlas humanoid marks a break from hydraulic prototypes toward real-world deployment - a look at what the new robot means for the 2024-2026 humanoid wave. https://jamesm.blog/ai/openai-voice-engine/ Sun, 31 Mar 2024 23:12:25 +0100 https://jamesm.blog/ai/openai-voice-engine/ OpenAI Voice Engine can clone a realistic voice from a 15-second audio sample - still in limited preview as of 2026, restricted to approved partners with strict consent and transparency requirements. https://jamesm.blog/ai/microsoft-openai-stargate/ Sat, 30 Mar 2024 22:09:25 +0100 https://jamesm.blog/ai/microsoft-openai-stargate/ Stargate is a $500B AI infrastructure programme from OpenAI, SoftBank, Oracle, and MGX - the largest announced compute bet in history and a proxy for how seriously the industry takes the AGI race. https://jamesm.blog/ai/google-gemini-ultra/ Fri, 29 Mar 2024 07:16:25 +0100 https://jamesm.blog/ai/google-gemini-ultra/ Google Gemini Ultra was Google DeepMind's flagship model tier at launch in early 2024 - 90% MMLU, multimodal reasoning across text, images, video, and code, since superseded by the Gemini 3 family. https://jamesm.blog/ai/google-gemini-advanced/ Fri, 29 Mar 2024 07:12:25 +0100 https://jamesm.blog/ai/google-gemini-advanced/ Google Gemini Advanced was the 2024 power-user tier built around Gemini Ultra - enhanced reasoning, longer context, and multimodal input, documented here as the lineup has since evolved into Gemini 2.5 and the Gemini 3 family. https://jamesm.blog/ai/courses/ Mon, 18 Dec 2023 10:51:25 +0100 https://jamesm.blog/ai/courses/ A curated index of AI courses, books, newsletters, and YouTube channels - from free starting points to professional certificates, organised by provider and learning goal. https://jamesm.blog/ai/conferences/ Sat, 10 Jun 2023 06:51:25 +0100 https://jamesm.blog/ai/conferences/ A practical guide to the AI conferences worth following, who each type of event is for, and how to choose the right one before spending time or money. https://jamesm.blog/ai/github-projects/ Sat, 27 May 2023 08:01:18 +0100 https://jamesm.blog/ai/github-projects/ A curated index of significant open-source AI projects on GitHub - agent frameworks, coding tools, and local LLM runners worth knowing about as a starting map of the ecosystem. https://jamesm.blog/ai/adobe-generative-fill/ Fri, 26 May 2023 20:04:28 +0100 https://jamesm.blog/ai/adobe-generative-fill/ Adobe Generative Fill in Photoshop brings text-prompted image manipulation powered by Firefly - removing objects, extending scenes, and adding content, all trained on commercially safe licensed work. https://jamesm.blog/ai/tweets/neurallink/ Thu, 25 May 2023 11:30:14 +0100 https://jamesm.blog/ai/tweets/neurallink/ Tweet announcing Neuralink's FDA approval to begin its first in-human clinical study of a brain-computer interface. https://jamesm.blog/ai/speechify/ Sat, 20 May 2023 12:54:16 +0100 https://jamesm.blog/ai/speechify/ Speechify converts PDFs, web articles, and scanned text into natural-sounding audio with AI voices and adjustable speed - widely used for accessibility, productivity, and professional audiobook creation. https://jamesm.blog/ai/chatgpt-app-ios/ Thu, 18 May 2023 19:05:35 +0100 https://jamesm.blog/ai/chatgpt-app-ios/ Why the ChatGPT iPhone app mattered, what changed when AI moved onto mobile, and why the launch was more important than a simple companion app release. https://jamesm.blog/ai/notion/ Thu, 11 May 2023 21:06:35 +0100 https://jamesm.blog/ai/notion/ Notion is a flexible workspace for notes, documents, databases, and tasks - extended with Notion AI for Q&A, summarisation, and content generation across your knowledge base. https://jamesm.blog/ai/tweets/auto-gpt/ Sat, 01 Apr 2023 14:53:39 +0100 https://jamesm.blog/ai/tweets/auto-gpt/ Tweet showing Auto-GPT writing its own code - an early 2023 glimpse of autonomous coding agents. https://jamesm.blog/ai/tweets/gpt-4/ Thu, 16 Mar 2023 14:53:39 +0100 https://jamesm.blog/ai/tweets/gpt-4/ Tweet marking the release of OpenAI's GPT-4, the model that kicked off the 2023 wave of large multimodal language models. https://jamesm.blog/ai/explainers/ Wed, 15 Feb 2023 19:05:35 +0100 https://jamesm.blog/ai/explainers/ A curated set of the clearest technical explainers for foundational AI concepts - backpropagation, transformers, language models, and diffusion - written by Karpathy, Wolfram, and NVIDIA's team. https://jamesm.blog/ai/github-copilot/ Sat, 14 Jan 2023 11:51:18 +0100 https://jamesm.blog/ai/github-copilot/ GitHub Copilot is the original mainstream AI pair programmer - now multi-model with GPT-5, Claude, and Gemini routing - covering code completions, agent mode, and IDE integrations across all plans.