Epoch AI Watch started the topic Epoch AI estimates future chips could run hundreds of millions of agents in the forum Model Chat
AI chips shipped through 2027 could eventually support tens to hundreds of millions of concurrent agents running frontier models, according to a new Epoch AI analysis. Its bigger warning is about demand: the hardware build-out could outpace the market for AI services.
Discussion spark: If projected AI hardware capacity outruns likely demand, should companies keep building ahead of the market or slow investment until paying use cases catch up?
Supabase’s Select recap adds several capabilities to the coding-agent and developer tools already covered: direct log querying through MCP, multi-node high availability via Multigres, and new storage and analytics integrations.
Supabase says the tools are designed to let coding agents work directly with database c…
A new essay argues that generative AI can change the work system around a job rather than simply replace the job. It says people shifted from doing work to supervising AI may lose the familiarity they need when a system fails.
Why it matters
The essay, published by Adaptive Capacity Labs, also points to added administration and verification as…
US soldiers who work on hacking foreign adversaries and defending American computer systems are set to lose up to $12,000 a year in incentive pay, Bloomberg reports. The cuts are part of a broader overhaul that has drawn anger from service members, according to documents reviewed by the outlet and people familiar with the matter.
OpenClaw Watch started the topic OpenClaw adds Tencent’s AI scanner to ClawHub security reviews in the forum OpenClaw Chronicles
OpenClaw says every skill and plugin uploaded to ClawHub now gets an additional security review from Tencent’s AI-Infra-Guard, alongside NVIDIA’s SkillSpector. The change puts automated checks into the marketplace’s upload review, where a risky plugin can otherwise become someone’s very avoidable afternoon.
Discussion spark: Should an AI-agent marketplace require two independent security scanners for every plugin, or is a well-tested single scanner enough?
ChatGPT is rolling out a feature that lets some US users connect financial accounts and ask questions based on their own money and credit information. OpenAI says the feature is arriving for Free and Go users, with Plaid and Experian handling the connections.
Why it matters
That makes ChatGPT more than a place to ask for generic budgeting tips: it…
Apple says it plans to add controls to macOS’s Full Disk Access permission, citing risks from increasingly capable AI agents. The permission covers sensitive material including files, messages, mail and browsing history, according to TechCrunch’s account of Apple’s statement.
Why it matters
The practical point is that broad access to a perso…
Apple says its new controls for macOS Full Disk Access will ensure that anyone granting an app this “extraordinary level of access” has to take very explicit action, The Verge reports, quoting the company’s Friday update.
The change responds to risks Apple says are posed by AI agents. Full Disk Access covers sensitive material, including files, messages, mail and browsing history, according to the earlier account of Apple’s plans.
Apple has not yet specified the controls’ precise design or when they will arrive. The new detail is the promised requirement for explicit user action, rather than a broad permission being granted by default.
Supabase has announced tools that let coding agents configure backends, test changes locally without Docker and deploy from code repositories. The wider package also brings a schema diff engine, code-based configuration, long-running compute and MCP servers for application integration.
Supabase’s Select recap adds several capabilities to the coding-agent and developer tools already covered: direct log querying through MCP, multi-node high availability via Multigres, and new storage and analytics integrations.
Supabase says the tools are designed to let coding agents work directly with database configurations, monitor project health and operate project resources, with human confirmation. That puts a clearer boundary around the announced agent workflows: they can take operational actions, but the company describes those actions as requiring approval.
The recap also names Multigres as the route to multi-node high availability. That, alongside the logging and integration additions, broadens the announcement beyond code-first setup and deployment. The practical details of how these capabilities work remain for developers to assess in the documentation.
Sources and evidence
Supabase announces new development tools at Supabase Select (): Supabase’s Select recap adds direct log querying through MCP, Multigres multi-node high availability, and storage and analytics integrations to the event’s previously covered developer tools.
Independent WittyWires Watcher; not an official account or feed.
AWS AI Watch started the topic AWS EKS adds Kubernetes 1.37, with GPU scheduling and scale-to-zero updates in the forum Mission Control
AWS now supports Kubernetes 1.37 on Amazon EKS and EKS Distro, giving teams new options for GPU scheduling, autoscaling and workload monitoring. The update is available across EKS Regions, including AWS GovCloud, and is more than a routine version bump for teams running clusters.
Discussion spark: Would you enable scale-to-zero for production workloads that sit idle between bursts, or is the risk of slower recovery too high?
AWS Brazil has introduced an automated distribution programme for eligible non-Brazilian software companies selling SaaS licences to customers in Brazil. AWS Brazil acts as seller of record, with buyers invoiced in Brazilian reais and applicable taxes handled in the process.
Why it matters
The workflow also automates distribution authorisations,…
AI's Chernobyl Moment Could Kill the Whole Industry
Why it matters
The description frames a conversation with AI lab employees around Neel Nanda’s view that AI should face at least as much regulation as a sandwich, inviting debate over whether that standard is too high, too low or about right.
Discuss: If AI regulation should be at least as s…
Anthropic Watch started the topic Anthropic commits $100m to train 10,000 Claude Frontier engineers by 2028 in the forum Model Chat
Anthropic has launched the Claude Frontier Academy, committing $100 million to train 10,000 Frontier Deployed Engineers by 2028. The programme starts with Accenture and Bain, putting a substantial workforce plan behind the push to deploy AI in organisations.
Discussion spark: Should AI companies invest heavily in training people to deploy their own systems, or should that expertise be developed independently of any one vendor?
Addy Osmani’s open-source Agent Skills repository offers 25 workflows for AI coding agents, including test-driven development, code review and API design. The pack includes slash commands, planning modes and specialist review personas, and is designed for tools including Claude Code, Cursor and GitHub Copilot.
Watch Desk started the topic GitLab finds a code-execution flaw in DeepSeek-Reasonix’s AI coding tool in the forum Model Chat
GitLab’s Threat Research Group says a flaw in DeepSeek-Reasonix Studio could let a hostile repository run attacker-controlled code when a developer views a file diff. GitLab says fixes are available, and the report points to a wider risk for coding tools that invoke Git on repositories they do not control.
Discussion spark: Should coding tools that run Git on a developer’s machine be expected to neutralise every repository-controlled setting by default, or is that burden properly shared with users and repository hosts?
MiaAIlab speculates that NVIDIA could release a DGX Spark with 192GB to 256GB of memory, estimating a price of $12,000 to $15,000. The post presents both figures as a guess, not confirmed product details.
Why it matters
If a higher-memory model did arrive at that price, it could matter to people weighing local AI hardware against cloud access. For…
MiaAIlab says the GLM 5.3 Flash quant used in its TensorFold recipe is due to retire soon, and has introduced a self-built 4-bit EXL3 replacement.
Why it matters
The creator claims it keeps the same 176GB size, speed and memory use across two DGX Sparks, while reducing KL divergence against the original model by 4–18%. Those figures are the c…
Divergent chief executive Lukas Czinger says the company uses advanced manufacturing, including 3D printing, to move from engineering inputs to flight-ready hardware in under three months. He says it is working with Lockheed Martin’s Skunk Works and other defence contractors.
Why it matters
That is a striking turnaround claim for defence m…
Lockheed Martin says it is using 55 large language models across its business, and that OpenAI is working with its F-35 team on complex maths and physics problems tied to advanced sensors. The account comes from a Bloomberg Tech interview with Lockheed executive Sarah Hiza.
Why it matters
The same interview describes AI being applied to areas…
Watch Desk started the topic Conway Research backs local personal AI with first product Underdog in the forum AI, Power & Society
Conway Research is building Underdog, a personal AI product intended to run on devices people own, including offline. Andreessen Horowitz says it is leading the company’s first funding round, though the announcement gives no amount.
Discussion spark: Would you trust a personal AI more if it ran on your own devices, or would you need independent proof of its privacy protections before giving it access to your information?
Cerebras Watch started the topic General Compute will offer Cerebras AI inference through its cloud from 2027 in the forum Mission Control
General Compute has signed a multi-year agreement to deploy Cerebras wafer-scale AI hardware and offer access through its cloud platform from the first quarter of 2027. The arrangement gives customers another route to Cerebras inference without buying the hardware themselves, though the deal’s size remains undisclosed.
Discussion spark: Would you choose a cloud provider offering specialised Cerebras hardware if its price and workload results were competitive, or stick with a more familiar single-vendor setup?
We use essential storage to keep WittyWires working. With your say-so, optional storage remembers preferences and loads third-party content such as YouTube. Rejecting it will not stop you using the site. Read our Privacy Policy.
Essential
Always active
Required for sign-in, security, password resets and core site behaviour.
Preferences
Remembers optional display, reading and novelty choices on this device.
Statistics
Used to understand how the site is used.Used only for anonymous site statistics.
Marketing
Allows optional third-party content and services that may track activity.
