01 · Money
Why one fintech keeps AI chatbots away from money decisions
Many people in the UK cannot afford a traditional financial adviser.
Daily brief · October 14, 2026
AI news without the tech headache.
Top story: Why one fintech keeps AI chatbots away from money decisions
Daily Brief
Moirangthem Margareth DeviStart here
AI in 30 seconds
02 · Chips
Applied Materials opens $5 billion chip research center in California
Chips power everything from phones to AI, and making better ones keeps getting harder and more expensive.
03 · Tools
AMD releases Ross, an AI helper for chip designers
Chip design is slow, skilled work, and faster tools mean cheaper and quicker electronics for everyone.
Top stories
What changed today.
FF News
Why one fintech keeps AI chatbots away from money decisions
The briefing:
Tom Richards, the technology chief at Aida by MoneyMeans, published a paper on building safe AI financial advisers. His argument: chatty AI models give incomplete answers on money questions. So the AI should only handle the conversation, while fixed rules written by licensed planners make the actual decisions.
Key details:
- Mouth and brain The AI model is the mouth that chats. A separate layer of checked, recorded rules is the brain that decides what advice to give.
- Humans in charge Regulated financial planners write and watch over the decision rules, so advice stays sound even if the chat model is replaced.
- Knows when to stop If the system lacks information or gets asked something outside its limits, it hands the customer to a human for review.
- Test the answers Richards says testing should check the calculations and legal outcomes, not just whether the conversation sounds smooth and friendly.
WHY IT MATTERS
Many people in the UK cannot afford a traditional financial adviser. Systems like this could offer regulated advice for less money. Think of it like a bank teller who can chat freely but must follow a strict rulebook before moving your money. That split may become the standard wherever mistakes carry legal risk.
Fortune
Applied Materials opens $5 billion chip research center in California
The briefing:
Applied Materials, the biggest maker of chipmaking machines, opened its EPIC research center in Sunnyvale, California. Intel, TSMC, Micron, SK Hynix and Samsung will all station researchers there. The company says working side by side could cut the time to bring a new chip design to market from about 10 to 12 years down to as little as six.
Key details:
- Massive building The center covers 600,000 square feet and used two Eiffel Towers' worth of steel. Total investment should reach $5 billion.
- Rivals share space Clean rooms the size of three football fields hold private test areas, so competing companies can run separate projects in one building.
- Early machine access Partner researchers get first use of Applied Materials' newest equipment, working next to the company's own engineers.
- Research only The site will not make chips for sale, so it does not by itself bring manufacturing back to the US.
WHY IT MATTERS
Chips power everything from phones to AI, and making better ones keeps getting harder and more expensive. Sharing a lab is like neighbors splitting the cost of expensive tools nobody can afford alone. If it speeds things up, new technology could reach your devices years sooner. The faster timeline is still a promise, not a proven result.
All About Circuits
AMD releases Ross, an AI helper for chip designers
The briefing:
AMD launched Ross, an AI assistant that helps engineers design chips and circuit boards. Unusually, it lets engineers plug in whichever AI model they prefer instead of forcing one choice. AMD says over 800 customers tried it during early access, and some design jobs that took weeks finished in two hours.
Key details:
- Bring your own AI Engineers can use any AI model and their usual coding apps, like VS Code or Cursor, rather than a locked-in tool.
- Works offline A fully disconnected mode, with AMD's documents stored locally, serves defense and safety-critical teams who cannot send data out.
- Big time savings claimed In AMD's demos, one image-processing design hit its speed targets in two hours instead of weeks, using about 35% fewer chip resources.
- Humans must approve The assistant cannot plan or act without a person reviewing and approving the step first.
WHY IT MATTERS
Chip design is slow, skilled work, and faster tools mean cheaper and quicker electronics for everyone. It is like a workshop welcoming any brand of power drill instead of selling its own. That openness pressures rivals like Intel to match it, because engineers rarely accept worse tools once they have tried better ones.
VnExpress
AI mortgage startup founded by university friends reaches $2.3 billion value
The briefing:
VnExpress reports that three founders with Harvard and Carnegie Mellon degrees built an AI mortgage startup now valued at $2.3 billion. The report shared few details beyond the valuation and the founders' backgrounds. Still, the number itself shows where investors think AI pays off first: slow, paper-heavy work like home loans.
Key details:
- Big valuation At $2.3 billion, the company ranks among the larger financial-technology outcomes of the current AI boom.
- Why mortgages Home loans involve clear documents, clear rules and expensive human work in the middle, a mix AI handles well.
- Few details shared The report did not include revenue, customer numbers or investor names, so those remain unknown.
WHY IT MATTERS
Getting a mortgage means weeks of paperwork and waiting. If AI can read and check loan files faster, buying a home could get quicker and maybe cheaper. Investors clearly believe this, and a valuation this size will attract many copycat startups pitching the same idea to lenders.
SiliconANGLE
Pluto Security raises $42 million to track AI agents inside companies
The briefing:
Pluto Security, a Tel Aviv startup, raised $42 million to help companies find and control AI agents, the software helpers that act on their own. The company says businesses often do not know how many agents run on their systems. Pluto maps them automatically and flags risky behavior before it causes harm.
Key details:
- Finds hidden agents The software scans employee computers, cloud systems and work apps to list every AI agent, noting which model and tools each one uses.
- Stops risky moves When an agent tries something dangerous, Pluto sends it a written instruction to stop. Companies set the rules for what is allowed.
- Free risk list A free service called Plutonium shares safety data on more than 4,500 public agent add-ons and connectors.
- Strong backers The money came from TLV Partners, Jump Capital, and a fund tied to a Check Point co-founder.
WHY IT MATTERS
As companies hand more tasks to AI agents, nobody wants an unknown helper quietly touching sensitive data. It is like discovering strangers have keys to your office. One caution worth noting: asking a misbehaving agent to stop is a polite request, and attackers may try to get around exactly that.
NVIDIA Blog
NVIDIA preps apps and a powerful open model for Spark PCs
The briefing:
NVIDIA says its software is ready for RTX Spark, a new class of powerful AI PCs, with laptops arriving Friday, October 16. The company announced easy setup for popular AI assistants, safety sandboxes that keep agents contained, and a new model called Nemotron 3.5 Bolt 74B that is fully open, training data included.
Key details:
- Agents come first Assistants like GitHub Copilot and Perplexity get simple setup on Spark, plus protected spaces so they cannot touch things they shouldn't.
- Smart switching NVIDIA's free Switchyard software decides whether a task runs on your own machine or in the cloud.
- A big open model Bolt 74B is a large model compressed to fit in 24GB of memory, released October 15 with its training data and recipes public.
- Faster creative apps Adobe's Photoshop and Premiere were rebuilt for up to double the speed, and NVIDIA says local AI downloads run five times higher than last year.
WHY IT MATTERS
Running AI on your own computer means more privacy and no monthly cloud bill for many tasks. NVIDIA wants Spark to be the natural home for AI helpers before competitors catch up, like stocking a new store before the neighbors open theirs. If it works, powerful personal AI gets much easier for regular people.
Workflow of the day
Connect Your LLM to AMD Toolchains
Lets your preferred AI model drive AMD's chip design tools directly.
Ross · Model Context Protocol · Vivado · Vitis HLS
Preferred LLM
your chosen model
Vivado project
FPGA design files
Vitis HLS code
synthesis sources
Ross
linked via Model Context Protocol
- routes prompts to toolchain
- drives Vivado commands
- runs Vitis HLS flows
Agent-run design task
one toolchain job done
Versus manual flow
- iteration time
- manual steps
- errors caught
Step by step
Choose model
Connect your preferred LLM through Model Context Protocol
Attach Vivado
Point Ross at an existing Vivado project
Attach Vitis
Link Vitis HLS for high-level synthesis work
Run task
Ask Ross to execute one real toolchain job
Compare
Measure Ross output against your current manual flow
Your model drives AMD toolchains through one open protocol.
Starter prompt
I am an embedded developer on AMD platforms (FPGA, adaptive SoC, embedded x86, or edge AI). I want to set up Ross, AMD's agentic AI assistant, this week. Walk me through: 1) connecting my preferred large language model to Ross through its open-standard Model Context Protocol servers, 2) attaching Ross to my existing Vivado project, 3) adding Vitis HLS for high-level synthesis tasks. Then suggest one small, low-risk toolchain task I should hand to Ross first, such as running a build step or querying synthesis results, and a simple way to compare the outcome against my current manual workflow on time taken, steps removed, and errors caught. Ask me which AMD platform and which LLM I use before you start.
One thing to try today
List every AI tool you actually use at work or home
Write down each AI assistant or tool you have used this month, what it can access, and what you would never let it touch. Companies pay millions for this kind of inventory. Yours takes ten minutes and shows you where your own risks sit.
~10 min
Learn from our free tutorial
Browse all tutorials
Featured tutorial
Gemini tutorial for beginners: research, images and music in three workflows
You will use Gemini to find facts, make pictures and make music, one simple step at a time.
Watch tutorialAI AGENTS DAILY
Understand less tech. Get more done.
A visual brief for people who want the useful part of AI without spending the morning decoding AI news.