AI news6
Perplexity launches Personal Computer for Mac, turning a Mac mini into an always-on AI agent
On April 16, Perplexity rolled out Personal Computer for Mac to Max subscribers ($200/month), an agent that runs locally with full access to the Mac file system, Apple Mail, Calendar, iMessage, and native apps. Users invoke it by pressing both Command keys; it responds to text or voice and auto-displays quick actions across any app. Pitched explicitly as a Mac mini always-on deployment for users who want a persistent machine doing agent work 24/7. Not available to $20/month Pro subscribers.
macrumors.com ↗Why it matters
First mainstream agent shipped as a local OS-integrated resident process rather than a cloud service, and the first explicit positioning of a consumer Mac mini as dedicated agent hardware.
Snap cuts 1,000 jobs — 16% of workforce — as CEO says AI generates over 65% of new code
Snap CEO Evan Spiegel announced April 15 that the company is laying off roughly 1,000 full-time employees and closing over 300 open roles, calling it a "crucible moment." Spiegel said AI now generates more than 65% of Snap's new code, more critical work is being handed to focused teams and AI agents, and the cuts will reduce annualised costs by more than $500M by H2 2026. Restructuring charges will run $95M–$130M in Q2. US employees get four months severance plus equity vesting and healthcare.
cnbc.com ↗Why it matters
First major public-company tech workforce cut where the CEO explicitly cites AI code generation share as the structural driver, not framing it as macro cost cutting.
Allbirds pivots into NewBird AI with $50M, converting the sneaker company into a GPU rental business
Allbirds announced this week a $50M financing deal and a corporate pivot into NewBird AI, retooling the footwear company to rent GPU compute capacity. The sneaker business is being wound down as the AI infrastructure play takes over the corporate shell. Marks a clean departure from sustainable consumer goods into pure AI infrastructure arbitrage.
therundown.ai ↗Why it matters
Sharpest example yet of a listed consumer-brand pivot into AI infra to access compute demand rather than incremental improvements to a flat product category — a signal of how many dormant company shells may follow.
OpenAI commits more than $20B to Cerebras over three years, plus $1B to fund Cerebras data centres
OpenAI agreed to pay Cerebras more than $20B over the next three years for servers powered by Cerebras's wafer-scale chips, with a possible equity stake on top. OpenAI will also provide Cerebras roughly $1B in direct co-investment to help fund new data centre buildout. This diversifies OpenAI's compute stack beyond Nvidia-first dependency and comes alongside the company's closed $122B round anchored by Amazon, NVIDIA and SoftBank at an $852B valuation.
reuters.com ↗Why it matters
Largest non-Nvidia AI compute deal on record and the strongest signal yet that frontier labs are actively building diversification into wafer-scale alternatives.
Sabi unveils a brain-reading beanie aimed at translating thoughts to text at 30 words per minute
Silicon Valley startup Sabi this week unveiled a non-invasive wearable — a beanie with 70,000 to 100,000 miniature EEG sensors woven into fabric — that translates internal speech into typed text via a "brain foundation model" trained on 100,000 hours of neural data from 100 volunteers. Target throughput at launch is 30 words per minute, slower than a keyboard but improving with continued use. First product is a winter beanie slated for late 2026, with a streamlined baseball-cap form factor planned after.
welcome.ai ↗Why it matters
First serious non-invasive BCI pitched as a consumer wearable rather than a medical device, and a direct challenge to Neuralink's surgical-implant path.
OpenAI GPT-5.4-Cyber ships as a dedicated security-reasoning variant
OpenAI released GPT-5.4-Cyber this week, a variant of GPT-5.4 specialised for cybersecurity workflows — vulnerability detection, reasoning over exploit chains, incident response triage — explicitly positioned against Anthropic's Claude Mythos in the security-model tier. OpenAI is framing it as the answer to enterprise security buyers who want a dedicated cyber model rather than a general-purpose one. Pricing and availability gated to enterprise tier at launch.
therundown.ai ↗Why it matters
Cyber has become the first clean vertical split in frontier-model product strategy, and both labs now ship domain-specialised variants instead of one-size-fits-all flagships.
AI in the nonprofit sector4
American Red Cross builds Clara AI — a multi-agent crisis response platform on AWS AgentCore
The American Red Cross is building Clara AI, a multi-agent platform powered by Amazon's AgentCore, to orchestrate specialised agents for disaster relief, blood services, military family support, and lifesaving training with real-time multilingual assistance at scale. The project was selected in the "Pathfinder – Generative AI" category of the 2026 AWS Nonprofit Imagine Grant, with funding, AWS credits, and hands-on support from the AWS Generative AI Innovation Center. Currently in prototype and testing, focused on preserving human connection in crisis response while freeing responders for complex work.
aws.amazon.com ↗Why it matters
Largest-named nonprofit multi-agent deployment to date, and the most concrete proof point for AgentCore-on-AWS as a framework for frontline humanitarian AI rather than enterprise workflow automation.
Singapore launches the National AI Impact Programme — 10,000 enterprises and 100,000 workers by 2029
On March 2, Singapore's Ministry of Digital Development and Information announced the National AI Impact Programme (NAIIP), committing to support 10,000 enterprises to adopt AI over the next three years and to train 100,000 workers in AI skills by 2029. The programme builds on Singapore's National AI Strategy 2.0 and will publish training partners, application processes, and course details in the first half of 2026. Singapore is positioning this as the single-country benchmark for nationwide AI adoption at SME scale.
imda.gov.sg ↗Why it matters
The most concrete government-led AI adoption target in APAC, and the programme nonprofits and capacity-builders in the region will be benchmarked against when bidding for delivery roles.
Google.org + AVPN APAC AI Opportunity Fund hits 49 grantees, with India training survey showing 63% improved AI understanding
The AVPN-administered AI Opportunity Fund, seeded by Google.org with $15M and expanded by another $12M in late 2025, has now funded 49 APAC nonprofits across 14 countries to run AI workforce upskilling. A post-training survey of 430 trainees by India's Head Held High Foundation found 63% reported improved understanding of generative AI and 55% agreed the training made AI opportunity fairer. The fund covers Australia, India, Indonesia, Japan, Malaysia, New Zealand, Pakistan, the Philippines, Singapore, South Korea, Taiwan, Thailand, and Vietnam.
avpn.asia ↗Why it matters
Concrete efficacy data from the largest AI-workforce-upskilling philanthropic programme in APAC, and the clearest read on how reliable short-cycle AI training actually is at moving comprehension.
Nava Labs and Miracle Foundation's ThriveWell push caseworker AI into live county-level child-welfare operations
Nava Labs is partnering with Riverside County in California to deploy generative AI that reduces administrative burden on frontline caseworkers and scales individual caseworker capacity. Separately, Miracle Foundation's ThriveWell platform — built on AWS AI services — is now in active use by caseworkers in child welfare to identify critical patterns, predict outcomes, and compress response times for at-risk families. Both are live, not pilots; both focus on the administrative-overhead problem that blocks more direct family time.
mcgovern.org ↗Why it matters
Caseworker AI has moved past vendor announcements into active county-scale deployment, and the blueprint is now concrete enough for other states and child-welfare NGOs to copy rather than reinvent.