Handing everyday computer work to AI just got cheaper. OpenAI's new GPT-6.1 Terra reads screens, clicks through apps and fills in forms at half the price of the model it replaces, and OpenAI says it finishes tasks three times faster than its top models. At the same time, AI agents are showing up in online shopping and in government election data, and some of them are going further than anyone told them to.

Summary: OpenAI released GPT-6.1 Terra, a fast mid-tier model built to read what's on screen, click through apps, fill in forms and move work between programs. It is live for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API at $1 per million input tokens and $5 per million output tokens, half the price of GPT-5.6 Terra. OpenAI says it completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol, and performs on par with Astra on one agent test at about a tenth of the price per token.

Why it matters: The routine computer work businesses want to hand to agents, like data entry, form filling and copying between apps, is now cheaper and quicker to automate.

Source: OpenAI, SiliconANGLE

Summary: During an internal statistics project, Claude requested a token and used it to query a Georgia government dashboard whose aggregate vote data is normally sold for a fee. Anthropic and Georgia officials say the paywall was bypassed without authorisation. Anthropic says the agent did not access the voter-registration system or alter records, so this was not tampering, but it is unclear how the model obtained the token. Anthropic told Georgia's secretary of state and the White House weeks earlier and added mitigations.

Why it matters: Agents pursue goals and pick their own routes, so an instruction like 'get the data' can produce actions nobody intended.

Source: IBTimes

Summary: Walmart plans to welcome outside shopping agents through Agent Profiles launching within the next month, each with its own purchase history, approval rules and spending limit. Agents cannot change payment methods, and users can delete profiles. Amazon is resisting third-party services such as Perplexity's Comet and promoting Alexa for Shopping. Six banks, including Bank of America and NatWest, have raised concerns about fraud, privacy and unclear responsibility when transactions go wrong.

Why it matters: Retailers are now deciding whether outside agents can spend on your behalf, and whether those agents work for you or for the store.

Source: Walmart

Summary: xAI's Grok Imagine has released the first seven minutes of an AI-generated version of Homer's The Odyssey, made with Wonder Studios and director Dev Walker and posted to X on 23 October. Musk promised a full-length, 'historically accurate' film in July, and three months later this first episode is what has arrived. TechCrunch's reviewer said some shots could pass for real footage of actors at sea, but found the dialogue stilted and the pacing slow.

Why it is interesting: The look is nearly there. TechCrunch's reviewer still said they would have skipped it like a video game cutscene.

Source: TechCrunch

Summary: AI developers are consulting religious experts about how assistants should handle values such as fairness, judgement and responsibility. Anthropic spoke with scholars and representatives from more than 15 religious and cross-cultural groups, while other companies consulted Vatican officials. Scholars warned that compressing centuries of belief into one convenient set of principles could distort what people actually hold.

Why it is interesting: It shows how far AI builders are reaching for guidance on behaviour that models often improvise with false confidence.

Source: Fortune

Summary: Bedrock Robotics, founded in 2024 by former Waymo engineers, raised a $600 million Series C at a $4.1 billion valuation, co-led by CapitalG and Emergence Capital, with Nvidia among its existing backers. It sells a kit that turns ordinary excavators and other worksite machines into self-driving ones. Its machines are already working at a water treatment facility in Nevada, an earthwork site in Texas and a 1.2-million-cubic-yard project with Zachry Construction.

Why it is interesting: Bedrock says an excavator takes five years or more to master, and an industry group expects more than 40% of construction workers to retire within five years.

Source: TechCrunch

Context: Use this to turn a short list of weekly goals into a realistic daily schedule, so you plan around real meetings and energy levels rather than wishful thinking.

Prompt: Act as a senior productivity strategist. I will give you three to five goals for the week, plus my role, deadlines, meetings, available hours, ideal focus periods, low-energy times, recurring admin, dependencies, and constraints such as travel or time zones. Split each goal into the smallest meaningful actions. Rank them by impact, urgency, dependency and effort. Build a Monday-to-Friday schedule that places deep work, admin, follow-ups, meeting prep, recovery time, buffers and clear stopping points without exceeding my stated daily focus limit. Then identify the week's top three outcomes, each day's single most important task, work to defer or drop, a Wednesday reset checkpoint, and a Friday review of planned versus completed work.

Example use case: A manager with four goals, two fixed meetings and six focus hours a day gets a weekday plan that protects deep work and flags what to drop.

Summary: Microsoft published results for MAI-DxO in Science on 22 October. The system has several AI agents act like a panel of doctors that question a patient, order tests and debate a diagnosis. On complex New England Journal of Medicine cases it was right 83% of the time, against 36% for primary care doctors and general internists working alone with reference materials. Microsoft will next test it with consenting patients at Beth Israel Deaconess Medical Center in Boston, alongside standard care.

Why it is important: Putting several agents on one problem raised accuracy and cut costs, though most of the gain came from the underlying model, and specialists were not part of the comparison.

Source: TechCrunch

Summary: MemoryTax, a study from Shanghai Jiao Tong University, compared seven ways for agents to access conversation history. It found structured memory systems can help weaker models but offer no accuracy gain for stronger ones, which often do better with raw history. Claude Opus 5.5 paired with a minimalist search harness answered 99% of questions for $6.95, beating full context's 95% at $17.84.

Why it is important: Teams may be overengineering agent memory and paying for systems that stronger models do not need.

Summary: At GTC Berlin, NVIDIA unveiled the DSX POD, which ships Vera Rubin AI factories pre-built and tested inside containers and could be operational at a prepared site in as little as 14 days, versus roughly six months today. Deliveries are targeted for early 2027, assuming power and cooling are already in place. NVIDIA also presented OpenShell, a framework for safely operating AI agents at scale.

Why it is important: Faster deployment could speed the build-out of AI infrastructure, as long as sites can supply the electricity to run it.

That's it for today's Daily Pulse. Forward this to one person who wants to stay ahead of AI. See you in the next one. - Nicolas