AI news5
Meta launches Hatch, its first paid AI agent, at up to $200 a month
Meta is rolling out a paid AI agent called Hatch, a consumer-friendly take on the open-source OpenClaw that builds working tools from a plain-language description and can schedule appointments and send emails. Internal documents show a free tier plus a "Hatch Plus" subscription with five to ten times the usage limits, priced up to $200 a month, with a broader US launch planned for July. It puts Meta directly against OpenAI, Anthropic, Microsoft's Scout and Google's Spark in the agent race. Hatch is also slated to power Meta's coming smart glasses and AI pendant.
the-decoder.com ↗Why it matters
Meta charging for AI for the first time signals the whole sector is shifting from free chatbots to paid agents that do real work.
OpenAI plans a ChatGPT "superapp" overhaul ahead of its listing
OpenAI is reorganising to turn ChatGPT into an all-in-one productivity platform bundling coding tools, autonomous AI agents and deeper enterprise integrations, the Financial Times reported. The move chases sticky, high-margin enterprise contracts and puts OpenAI head-to-head with Anthropic, which has been winning corporate AI coding work. The redesign comes as OpenAI prepares a share listing after converting to a for-profit structure. Likely features include a native code editor, agentic workflows and admin controls for IT teams.
logicity.in ↗Why it matters
the major labs are converging on the same bet, that whoever owns the everyday work surface owns the enterprise.
Willow raises $7M to govern AI agents inside enterprises
Willow emerged from stealth with $7 million in seed funding led by Hetz Ventures to build a governance layer that controls how AI agents access internal company systems. The platform gives organisations visibility into agent activity, enforces what agents can touch and do, and flags unauthorised AI use, with a marketplace of more than 1,000 connectors. It already runs inside Wix across more than 5,000 employees and is expanding into cybersecurity, real estate and fintech. The founders are former Wix engineers.
pulse2.com ↗Why it matters
the gap between AI agents and an organisation's internal data is becoming the key security control, and the same governance question lands on any nonprofit deploying agents.
Google ships an "agentic RAG" framework that lifts factual accuracy by up to 34%
Google Research and Google Cloud released an agentic retrieval-augmented-generation framework on the Gemini Enterprise Agent Platform that breaks complex queries into steps and keeps searching until it has enough context before answering. Google reports up to a 34% accuracy gain on factuality datasets versus standard RAG, plus better grounding on internal company data. It is built so responses stay auditable and traceable, and is now in public preview. This is aimed squarely at enterprises that need dependable, sourced answers.
research.google ↗Why it matters
grounded, citable RAG is the difference between an AI a nonprofit can trust on real records and one that quietly makes things up.
Cegid buys Shine to build Europe's first all-in-one AI finance platform for SMBs
European software firm Cegid closed its acquisition of French fintech Shine, backed by a new €1.1 billion financing facility, to create what it calls Europe's first fully cloud, AI-driven financial platform for small businesses and accountants. The combined platform spans e-invoicing, accounting, payments, tax, payroll and reporting, positioning SMBs for the EU's mandatory e-invoicing reforms. Shine serves more than 400,000 customers across France, Denmark, the Netherlands and Germany.
morningstar.com ↗Why it matters
AI is being baked into the everyday finance tools small organisations already run, which is often where resource-strapped teams feel the first real time savings.
AI in the nonprofit sector3
An AI chatbot routes Phoenix residents to heat-relief centres around the clock
The nonprofit Valley Leadership has expanded Just Ask Mia, a free 24/7 anonymous AI chatbot, to route metro Phoenix residents to the nearest cooling centre as summer temperatures climb. Working with Maricopa County public health, Mia now holds more than 200 cooling centres and hydration stations plus over 1,000 verified resources for food, clothing and housing, and answers in English or Spanish by text or web chat. For anyone in crisis it hands off to 211 Arizona for live human support. Mia first launched in 2024 and keeps its database current through partnerships with local agencies.
kjzz.org ↗Why it matters
a lean nonprofit is using a simple chatbot to make life-saving services findable in seconds, a clean example of AI extending frontline reach without replacing people.
A new AI app lets occupational therapists assess homes for hazards remotely
Researchers have launched myAccessibleHome, an AI-assisted app that lets an occupational therapist assess a home for fall and accessibility hazards without an in-person visit. A caregiver answers a few questions and scans the home with a smartphone, augmented reality takes precise measurements, and the therapist reviews the report before recommending changes and connecting the household to local installers. It was built on more than a decade of research with $1.6 million in US federal disability-research funding, and the iOS version launched this month through partner organisations. The clinical decisions stay with the therapist, with AI handling speed, reach and data capture.
searchlightnm.org ↗Why it matters
it shows AI extending scarce specialist capacity into rural and under-served areas while keeping the human professional in charge of the call.
A foundation's Claude tool standardises grant due diligence for 100-plus funders
The Patrick J. McGovern Foundation built Grant Guardian, a Claude-powered tool that reads nonprofit financial documents in any format, extracts key metrics, applies each funder's evaluation criteria and flags areas needing a closer look. It now serves more than 100 philanthropies and has processed hundreds of grant applications, with United Way affiliates reporting significant time savings and the ability to run due diligence in-house. Data privacy was the deciding factor in choosing the model, given the sensitive financial records involved. The tool deliberately produces no single score, keeping human judgment at the centre of funding decisions.
theapplied.co ↗Why it matters
tools like this hand smaller funders the kind of rigorous, consistent analysis that used to be a big-foundation advantage, which over time shapes how money reaches the sector.