🏦 Your Bank Hired an AI

London startup Gradient Labs is giving every bank customer the equivalent of a dedicated account manager - powered by GPT-5.4 mini and nano. The company, founded by the team that previously led AI at Monzo, has built AI agents that handle real customer calls: verifying identity, freezing stolen cards, initiating replacements, and answering follow-up questions in under 500 milliseconds.

Key Insights:

• 500ms response time - fast enough for natural voice conversations, with GPT-5.4 mini and nano handling the heavy lifting

• 97% trajectory accuracy - meaning the AI follows the correct procedure from start to finish on complex banking calls, versus 88% for the next-best provider

• 15+ guardrail systems run in parallel on every call, checking for compliance violations, vulnerability signals, fraud attempts, and financial advice boundaries

• Gradual deployment - banks start with low-risk workflows and expand over time, with continuous monitoring flagging conversations that need human review

• Built by ex-Monzo AI leads - the founding team brings direct experience building AI systems inside a regulated digital bank

The Bigger Picture: Most banks still route customers through phone trees and transfer queues. Gradient Labs is betting that AI can deliver the kind of personalized service that used to be reserved for private banking clients - to everyone. The 97% accuracy figure matters because in banking, a wrong step isn't just a bad experience, it's a compliance incident. If this model works at scale, the question for every bank becomes: why are your customers still waiting on hold?

Read More →

💊 AI Now Prescribes Psychiatric Meds

Utah has approved a one-year pilot allowing Legion Health's AI chatbot to renew prescriptions for psychiatric medications - only the second time any US state has given this kind of clinical authority to AI. The $19/month subscription service can refill 15 lower-risk maintenance drugs, including Prozac, Zoloft, and Wellbutrin, for patients who are already stable on their treatment plan.

Key Insights:

• 15 medications approved - all lower-risk maintenance drugs already prescribed by a clinician, including fluoxetine, sertraline, bupropion, and hydroxyzine

• Strict guardrails - patients with recent dose changes, medication switches, or psychiatric hospitalization in the past year are excluded

• Red flag screening - the chatbot asks about suicidal thoughts, self-harm, severe reactions, and pregnancy before processing any refill

• Human escalation built in - if any answer falls outside low-risk criteria, the case goes to a clinician before any refill is issued

• 500,000 Utah residents currently lack access to mental health care, which state officials cite as the primary motivation for the pilot

The Bigger Picture: This pilot sits at the intersection of AI's biggest promise and biggest risk. On one hand, half a million people in Utah can't see a psychiatrist. On the other, Harvard psychiatrists warn the system could keep people on medications longer than they should be - automating refills without the nuanced judgment of whether someone actually still needs the drug. The narrow scope is designed to limit harm, but it sets a precedent that other states will be watching closely.

Read More →

🦾 Gig Workers Now Train Robots

Thousands of contract workers across 50+ countries are strapping iPhones to their heads and filming themselves doing household chores - folding laundry, washing dishes, cooking - to train the next generation of humanoid robots. Palo Alto-based Micro1 is paying workers $15/hour to record these videos, which are sold to companies like Tesla, Figure AI, and Agility Robotics, racing to build human-like machines.

Key Insights:

• $15/hour pay - good income in markets like Nigeria and India, where Micro1 recruits tech-savvy workers through LinkedIn and YouTube

• $6 billion invested in humanoid robots in 2025 alone, with at-home data recording becoming a booming global gig economy

• $100 million+ spent annually by robotics companies buying real-world movement data from Micro1 and competitors like Scale AI and Encord

• DoorDash pays drivers to film themselves doing household chores, while China operates state-run robot training centers with VR headsets and exoskeletons

• AI vetting process - an AI agent named Zara interviews applicants, reviews sample videos, and manages quality control across the global workforce

The Bigger Picture: Just as ChatGPT learned language from billions of web pages, humanoid robots need to learn movement from real humans in real homes. Simulations can teach a robot to backflip but not how to fold a shirt - physics is too messy to model perfectly. What's emerging is a new global labor market where people in developing economies earn decent wages teaching robots to do the very tasks those robots may eventually replace. It's a strange loop, but the demand for human movement data is growing faster than anyone expected.

Read More →

💰 OpenAI Raises $122B at $852B

OpenAI closed its largest funding round ever - $122 billion at an $852 billion valuation - with SoftBank, Amazon, Nvidia, and Microsoft all participating. The company now generates $2 billion in revenue per month, has 900 million weekly active users, and is widely expected to IPO this year.

Read More →

🧬 Anthropic Buys Biotech for $400M

Anthropic acquired Coefficient Bio, a biotech startup, for $400 million - signaling a push beyond chatbots into scientific research and drug discovery. The acquisition comes as AI labs increasingly compete to prove their models can accelerate real-world science.

Read More →

🎙️ Microsoft Ships Three New Models

Microsoft released three in-house MAI foundational models that transcribe voice to text, generate audio, and create images - the first major output from a team formed just six months ago. The move signals Microsoft is building its own AI capabilities alongside its $13 billion OpenAI partnership.

Read More →

🗺️ Google Maps Gets AI Upgrade

Google Maps launched Ask Maps, a conversational AI feature powered by Gemini that answers complex questions like finding EV chargers near landmarks or planning multi-stop road trips. The update rolled out globally alongside a redesigned navigation experience in over 200 countries.

Read More →

Alibaba Accio 🛒

Alibaba's AI sourcing tool hit 10 million monthly active users in March - meaning one in five Alibaba users now consults AI when finding manufacturers. Small sellers describe product ideas in plain language and Accio suggests design changes, identifies factories, and cuts sourcing timelines from months to days. One seller reduced manufacturing costs from $17 to $2.50 per unit.

Try it →

The Neuron's AI Proficiency Guide 📈

A practical 5-level framework for getting real value from AI in 2026: Projects, Prompting, Skills, Automations, and Agents. The guide breaks down which tools - ChatGPT, Claude, Gemini, Grok - work best for what, and explains why most people are stuck at Level 2 when the real gains live at Levels 3 through 5.

Try it →