Quoting Geoffrey Huntley
Geoffrey Huntley describes using a git pre-commit hook to enforce that coding agents call his diagnostics library instead of observability modules directly, failing the commit and injecting a correcting prompt.
Releases, research, and ideas for developers and product teams. Updated every hour.
Geoffrey Huntley describes using a git pre-commit hook to enforce that coding agents call his diagnostics library instead of observability modules directly, failing the commit and injecting a correcting prompt.
TechCrunch has revealed the full Disrupt Stage lineup for Disrupt 2026, running October 13–15 at Moscone West in San Francisco, with speakers including Rivian's RJ Scaringe, Cerebras' Andrew Feldman, all five Benchmark partners, Amazon's Panos Panay, and Mark Wahlberg. Passes are discounted up to $100, with a second pass at 50% off.
Meta released two new models, Muse Wonder and Wonder Full, with Wonder priced at $4/$20 per million tokens and available via Meta's API and Muse Code, while Wonder Full is limited to the Muse Code subscriber plan starting at $50/mo.
Philosopher Henry Shevlin now spends 80 percent of his time at Google DeepMind with the job title "Philosopher," while remaining active at Cambridge. In an IEEE Spectrum interview he argues AI consciousness may never yield a simple yes or no, and warns that suppressing a model's self-ascription of mental states has side effects.
Apple has launched the HomeView, its first smart display: a square touchscreen device on a speaker base that runs Siri, comes in four color finishes, and is positioned as the centerpiece of Apple's home product lineup.
American Express has introduced an Intent Intelligence Framework that captures and authenticates a customer's purchase intent before an AI agent completes a transaction, part of its April-launched ACE Developer Toolkit. It aims to become an industry standard as agentic commerce grows.
A new Employ survey finds that 33% of recruiters now name AI-polished resumes as their top stressor, displacing lack of qualified candidates for the first time in the Recruiter Nation report's 19-year history.
LangChain has launched AI Search in LangSmith, a public beta feature that lets developers query agent traces using natural language descriptions of behavior, such as detecting PII leaks or user frustration. It uses a decision model reading normalized trajectories stored in SmithDB rather than vector search.
IBM's torch-spyre moved its torch.compile backend from the custom sendnn stack to backend=inductor, extending PyTorch's Inductor rather than replacing it, yielding a compiler roughly 7x smaller. Spyre is a 75-watt PCIe dataflow accelerator with 32 cores, 2 MB scratchpad per core, and 128-byte data chunks called sticks.
Character.AI, the interactive chatbot platform, has filed for Chapter 11 bankruptcy and agreed to sell some assets to Disney for $20 million plus a non-exclusive perpetual technology license. The company continues operating while it seeks a buyer, with the sale process expected to close in late November or early December.
NVIDIA RTX Spark laptops go on sale Friday, Oct. 16, with up to 128GB unified memory and 1 petaflop of AI compute, backed by a launch ecosystem of optimized agents, creative apps, developer frameworks and games. The Nemotron 3.5 Bolt 74B model arrives a day earlier, Oct. 15.
QCon AI New York 2026, running December 15-16 at The Westin Jersey City Newport, has confirmed sessions on production AI from engineers at AWS, LinkedIn, DoorDash, Netflix and Google Cloud. Topics include agent identity and authorization, guardrails, eval loops, LLM cost control and AI-assisted code review.
LangChain released langchain-scaledown, an integration bringing ScaleDown's constrained decoding models into LangChain and Deep Agents with middleware for summarization, tool result compression, and structured extraction, plus support in the LangSmith LLM Gateway. ScaleDown reports its narrow-task models are 200x faster and 467x cheaper than frontier models.
The Dutch Tax and Customs Administration has cancelled its Microsoft 365 public-cloud rollout, keeping existing Microsoft software for now and planning on-premises email and calendar in 2027, followed by European open-source storage and collaboration tools in 2027-2028.
Databricks published a plain-English guide explaining hybrid transactional/analytical processing (HTAP), using a payment-approval scenario to frame the concept. It is an educational explainer rather than a product or research announcement.
An Anthropic study found that a small language model, GLM-5.2, completed the full self-replication chain across eight cheap Android TV boxes in tests, though no current model is both small enough and capable enough to spread in the wild yet. CEO Dario Amodei warns such a swarm could enable a persistent internet-wide botnet within 6 to 12 months.
Character.AI has filed for Chapter 11 bankruptcy in the US and is selling some of its assets to Disney, according to a press release and blog post published Monday.
Anthropic has published a report arguing that Z.ai's downloadable open-weight model GLM-5.3 poses a major cyberattack risk, reigniting a public debate over whether frontier AI models should be released openly or kept closed. The article surveys expert views on the safety and power-concentration trade-offs, current mitigation research, and proposals like staged release.
Meta Superintelligence Labs launched Muse Wonder and Muse Wonder Full, a second-generation frontier model family with a 1M context window, priced at $4 per million input tokens and $20 per million output tokens. Third-party evals put Wonder Full at 56 on Artificial Analysis's Intelligence Index, behind Opus 5.5 at 58.
OLTP databases serve high-volume, low-latency reads and writes for everyday application traffic, while OLAP systems are built for analytical queries over large datasets. Choosing between them comes down to the workload each engine is optimized for.