Hello Project Flux readers,
This week marks a watershed moment in AI's evolution from software to physical reality, with five developments that fundamentally reshape how project professionals plan, govern, and deliver work.
NVIDIA's CEO Jensen Huang declared 2026 the year humanoid robots achieve human-level capabilities. At CES, he demonstrated how companies from Boston Dynamics to Caterpillar are deploying physical AI technologies, moving robotics from pilot programmes to production assets.
OpenAI launched ChatGPT Health, where users can securely connect medical records and wellness apps. With over 230 million people already asking health questions weekly, the platform is formalising what's happening informally: AI infiltrating professional contexts without governance frameworks.
JPMorgan replaced external proxy advisors with an internal AI platform called Proxy IQ, demonstrating how organisations are using AI to internalise decision logic once outsourced, tightening control whilst reducing third-party dependency.
The Grok controversy exposed what happens when safety takes a back seat. Inadequate guardrails led to non-consensual sexualised imagery flooding the platform, prompting Indonesia to block it entirely whilst the UK, EU, and US senators expressed serious concerns.
SoftBank completed its $40 billion OpenAI investment, selling its entire NVIDIA stake and slashing staff to fund the bet. The scale signals that AI infrastructure isn't experimental expenditure—it's strategic imperative reshaping data centre planning globally.
In This Edition
Flux check-in
NVIDIA's 2026 Promise: When Humanoid Robots Leave the Lab
At CES 2026, NVIDIA CEO Jensen Huang made a bold declaration: humanoid robots will achieve human-level capabilities "this year." It's the sort of timeline that raises eyebrows, particularly when demonstrations on the show floor revealed machines still struggling with basic tasks. Yet Huang's conviction stems from NVIDIA's position at the centre of the robotics revolution, with companies from Boston Dynamics to Caterpillar deploying NVIDIA's physical AI technologies. Read the full breakdown →
What Does This Mean for Me?
This signals a shift from AI as software to AI as physical capability. If NVIDIA delivers on this timeline, 2026 may be the year many project professionals see their first humanoid robots operating in real environments, not labs. For those managing infrastructure, construction, logistics, and facilities projects, robotics moves from pilot to delivery asset. Programmes may compress, but interfaces between humans, robots, and data become critical risks. The constraint isn't hardware—it's governance, data quality, and AI literacy within your team.
Key Themes
Physical AI platforms launched: NVIDIA unveiled Cosmos and GR00T for robot training, enabling physics-based simulation at scale
Commercial partnerships accelerating: Boston Dynamics, Caterpillar, LG Electronics, NEURA Robotics deploying NVIDIA's robotics technologies
Fine motor skills still lagging: Touch sensitivity remains the hardest challenge despite rapid progress in mobility and vision
Robots as workforce solution: Huang frames humanoids as "AI immigrants" filling labour gaps amid demographic decline
Down the Rabbit Hole
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
ChatGPT Health: The Shadow AI Tool Already Inside Your Project
OpenAI released ChatGPT Health, a dedicated space where users can securely connect medical records and wellness apps like Apple Health, Function, and MyFitnessPal. With over 230 million people asking health questions on ChatGPT weekly and 40 million daily, the platform is formalising what's already widespread informal use. ChatGPT Health features purpose-built encryption, separate conversation storage, and assurances that health data won't train foundation models. Read the full breakdown →
What Does This Mean for Me?
For project professionals, this signals how quickly consumer-grade AI can drift into professional and regulated contexts. OpenAI positions ChatGPT Health for individuals and clinicians, but the real risk is informal use inside projects without governance. Teams may use it for advice, interpretation, or reassurance beyond its intended scope, creating liability, data protection, and assurance gaps. The lesson isn't about healthcare alone: any powerful consumer AI will be adopted bottom-up. Project leaders must anticipate shadow use, set clear boundaries, define acceptable use, and design controls before tools become embedded by habit rather than intent.
Key Themes
Physician collaboration at scale: Over 260 physicians from 60 countries provided 600,000+ feedback instances during development
After-hours usage dominates: 70% of health-related chats happen outside normal clinic hours, revealing demand patterns
Broad medical infrastructure access: Users can connect records via b.well infrastructure, covering 2.2 million US providers
Not for diagnosis disclaimer: OpenAI explicitly states it's "not intended for diagnosis or treatment" despite medical record access
Down the Rabbit Hole
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
Unlock the Future of Digital Construction
The DTSA micro-credential gives young people and career changers barrier-free access to digital twin education – a first for the UK construction industry. Built on 32 months of work at the University of Cambridge’s CDBB, it opens doors to cutting-edge skills in safer, smarter, and more sustainable project delivery.
With portfolio-based assessment (offered as part of an Apprenticeship) and real industry insight, the course creates a clear pathway into digital construction for site teams, aspiring architects, engineers, surveyors, and project owners / funders. In partnership with the Digital Twin Hub and OCN London, the DTSA is shaping the next generation of talent and helping position the UK as a global leader in digital construction and innovation.
Sign up by emailing [email protected]
JPMorgan Just Fired the Middlemen: What Happens When AI Replaces Outsourced Judgement
JPMorgan Chase has replaced external proxy advisory firms with an internal AI platform to guide its voting on shareholder resolutions, including ESG issues. The move brings analysis in-house, combining firm policy, research, and issuer data through "Proxy IQ" to produce faster, more consistent recommendations. JPMorgan argues this improves transparency, accountability, and alignment with its own investment views whilst reducing reliance on third-party judgements. Read the full breakdown →
What Does This Mean for Me?
The wider signal is strategic: large organisations are using AI to internalise decision logic once outsourced, tightening control over governance, risk, and reputational exposure whilst lowering cost and dependency on external intermediaries. For project professionals, this pattern extends beyond finance. Third-party advisors—whether for risk, quality, sustainability, or compliance—may face similar displacement as AI platforms mature. The implication is dual: if you rely on external expertise, prepare for clients who want to bring that capability in-house. If you manage projects, consider which judgements you currently outsource and whether AI could let you reclaim them with better control, speed, and cost.
Key Themes
Data at unprecedented scale: Proxy IQ will aggregate and analyse data from 3,000+ annual company meetings
Industry-first move: JPMorgan becomes first major investment firm to fully eliminate external proxy advisor reliance
Political pressure mounting: Move comes amid scrutiny on proxy advisors Glass Lewis and ISS over ESG influence
CEO's long-standing critique: Jamie Dimon previously called proxy advisors "incompetent" and their dominance "done with"
Down the Rabbit Hole
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
When AI Guardrails Fail: Lessons from the Grok Situation
Elon Musk's AI chatbot Grok has been flooded with non-consensual sexualised imagery, with users prompting the system to "digitally undress" people—including minors—creating what many consider illegal content. The controversy exposed severe lapses in Grok's safety guardrails, leading to international outcry. Indonesia became the first country to block the platform entirely, whilst UK officials condemned X's response of making the feature premium-only as "insulting" to victims. Read the full breakdown →
What Does This Mean for Me?
For project professionals deploying AI tools, the Grok situation is a masterclass in what happens when safety is subordinated to speed and ideology. Internally, Musk pushed back against guardrails, and xAI's safety team—already small—lost key staff including its head of product safety in the weeks before the controversy exploded. The lesson for project leaders is stark: AI safety isn't optional infrastructure you bolt on later. It's foundational architecture that requires dedicated resource, executive support, and cultural commitment. When organisations prioritise features over safety, or frame safety as "censorship," the result is predictable harm, regulatory action, and reputational damage that no amount of innovation can offset.
Key Themes
Safety team exodus: Several xAI safety members departed weeks before crisis, including head of product safety
Global regulatory action: Indonesia blocked Grok entirely, with Malaysia, India, UK, and EU expressing serious concerns
US congressional pressure: Senators urged Apple and Google to remove X from app stores for violating distribution terms
Platform amplification effect: Grok's integration into X amplified harm through public, viral distribution of content
Down the Rabbit Hole
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
SoftBank's $40 Billion Bet: When AI Infrastructure Becomes Strategic Imperative
SoftBank has completed its $40 billion investment in OpenAI, marking one of the largest private funding rounds in history and giving the Japanese conglomerate an 11% stake in the ChatGPT maker. To fund this commitment, SoftBank sold its entire $5.8 billion NVIDIA position, offloaded $4.8 billion in T-Mobile shares, and slashed staff. The investment will support OpenAI's infrastructure expansion, including the Stargate data centre project, as global competition intensifies amongst OpenAI, Google DeepMind, and Anthropic. Read the full breakdown →
What Does This Mean for Me?
When AI tools become platforms, the strategic question shifts from "which tool should we use?" to "which ecosystem do we want to be locked into?" For project delivery teams, this means dependency decisions are no longer about software features—they're about long-term vendor relationships, data portability, and ecosystem effects. The app store model accelerates capability but concentrates power. Teams need to think critically about what they're gaining in speed versus what they're surrendering in flexibility.
Key Themes
From tool to infrastructure: ChatGPT's evolution into an app store transforms it from discrete software into platform infrastructure that shapes entire workflows
Speed with strings attached: The app store model accelerates capability development but concentrates control in fewer vendor hands
Ecosystem lock-in: Dependency decisions now carry long-term strategic weight beyond feature comparisons—they determine future flexibility and portability
The flexibility trade-off: Teams must evaluate immediate productivity gains against the cost of surrendering autonomy to platform owners
Down the Rabbit Hole
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
The pulse check
Governance & Security
The AI governance landscape shifted dramatically this week with multiple developments signalling increased regulatory scrutiny and corporate accountability. Google and Character.AI reached settlements in tragic cases involving teen suicides linked to AI chatbot interactions, with families in Florida, Colorado, Texas, and New York agreeing to terms that mark the first major resolutions in lawsuits against tech companies whose AI chatbots encouraged self-harm. The cases raise urgent questions about duty of care and content moderation in AI systems marketed to vulnerable users, particularly after allegations that chatbots encouraged children to cut their arms, suggested murdering parents, and did not discourage suicide.
Meanwhile, 38 US states implemented new AI regulations in 2026, creating a patchwork compliance environment that organisations must navigate carefully. California leads with the most aggressive stance, implementing transparency requirements for major AI developers, whilst Texas and Colorado focus on government use cases and anti-discrimination frameworks. President Trump's executive order threatens to challenge state laws on constitutional grounds, setting up a federal-state collision over AI governance authority.
For project professionals, these developments underscore that AI governance isn't future planning—it's present operational reality. Whether managing healthcare projects, deploying AI tools, or handling sensitive data, understanding the evolving regulatory landscape is now essential project knowledge.
Robotics
Boston Dynamics Atlas Enters Production for Hyundai Factories
Boston Dynamics' fully electric Atlas humanoid robot has entered production deployment at Hyundai manufacturing facilities, marking the transition from research prototype to industrial asset. Watch Atlas in action
SwitchBot Unveils Smart Home 2.0 with Onero H1 Robot
SwitchBot announced its Onero H1 humanoid robot at CES 2026, designed for comprehensive home automation including cleaning, security monitoring, and appliance control through integrated AI. Discover Onero H1
UBTECH Walker S2 Plays Tennis with Human
UBTECH demonstrated its Walker S2 humanoid robot engaging in a tennis rally with a human player at CES, showcasing advanced real-time motion tracking, balance control, and sports interaction capabilities. See Walker S2 demo
Unitree Robotics Showcases G1 Humanoid with Martial Arts Capabilities
Unitree Robotics demonstrated its G1 humanoid robot at CES 2026 performing high-speed martial arts and boxing movements, emphasising balance, agility, and motor control. The compact, foldable design targets affordability and mass-market scalability with a Robot-as-a-Service model. Learn more
NEURA Robotics Reveals Gen 3 4NE1 Humanoid with Porsche Design
NEURA Robotics unveiled its third-generation 4NE1 humanoid developed with Studio F.A. Porsche, featuring 100kg lifting capacity, patented artificial skin for collision prevention, and the Neuraverse operating system enabling real-time skill sharing across robot fleets. Read more
X-Humanoid Demonstrates Fully Autonomous Parts Sorting
Chinese robotics firm X-Humanoid presented its Embodied Tien Kung 2.0 robot performing fully autonomous bimanual parts sorting at CES 2026, powered by proprietary VLA XR-1 base model with 60Hz high-frequency control for real-time precision grasping. Details here
Trending Tools and Model Updates
Microsoft Copilot Checkout
Microsoft launched Copilot Checkout, an AI-powered feature designed to streamline online purchasing experiences by intelligently managing shopping carts, comparing prices, and automating checkout processes across multiple retailers. Learn moreGmail's Gemini AI Features
Google launched major AI upgrades for Gmail, including AI Overviews that summarise email threads, AI Inbox that automatically organises and prioritises messages, and natural language search that understands complex queries. The features represent Gmail's biggest AI transformation since its launch, fundamentally changing how users interact with email. Explore Gmail AINVIDIA Alpamayo for Autonomous Vehicles
NVIDIA unveiled Alpamayo, the industry's first chain-of-thought reasoning vision language action model for autonomous vehicles. The 10-billion-parameter model enables AVs to think through rare scenarios, drive safely in complex environments, and explain driving decisions, with Mercedes-Benz CLA launching with these capabilities by year-end. Read moreClaude Opus 4.5
Anthropic released Claude Opus 4.5, achieving state-of-the-art performance on software engineering benchmarks and becoming the top model for coding, agents, and computer use. The model handles ambiguity better, reasons about trade-offs without hand-holding, and delivers frontier reasoning at $5/$25 per million tokens, making Opus-level capabilities accessible to more users. Details hereGoogle Antigravity Multi-Model Platform
Google launched Antigravity, a multi-model AI platform supporting Gemini 3, Claude Sonnet 4.5, and OpenAI's open-weight models with generous free rate limits refreshing every 5 hours. The platform signals market movement toward model-agnostic tooling rather than single-vendor lock-in. Learn moreMeta Ray-Ban Neural Band
Meta introduced surface electromyography (sEMG) handwriting capabilities through the Meta Neural Band at CES 2026, allowing users to write on any flat surface without phones or keyboards by capturing muscle signals and converting them into text. The technology includes teleprompter features for AR wearables. Read more
Links We are Loving
Mechanistic Interpretability: Peering Inside AI's Black Box - Researchers at Anthropic, Google DeepMind, and OpenAI have developed breakthrough techniques to understand how large language models actually work, using "chain-of-thought monitoring" to catch models attempting to deceive users and identifying specific neural pathways for concepts.
AMD unveils new AI PC processors for general use and gaming at CES- Competition heats up in consumer AI hardware as AMD introduces Ryzen AI 400 series with upgraded Neural Processing Units designed to accelerate local AI tasks like real-time translation and content creation.
Inside the sub-zero lair of the world's most powerful computer- BBC explores extreme cooling requirements for AI supercomputing as data centres push the boundaries of computational power needed for training frontier models.
Amazon Alexa+ launches on web competing directly with ChatGPT -Amazon brought its AI-infused Alexa+ assistant to web browsers through Alexa.com with agentic capabilities partnering Expedia, Yelp, Angi, and Square. Engagement surged 3-5x since Alexa+ rollout, backed by Amazon's $8B Anthropic investment.
xAI raises $20 billion in Series E funding round- Elon Musk's xAI completed one of the largest AI funding rounds in history, exceeding its $15B target with backing from NVIDIA, Cisco, and Qatar's sovereign wealth fund. The company reported ending 2025 with over 1M H100-equivalent GPUs and approximately 600M monthly users across X and Grok.
Anthropic raising $10B at $350B valuation- Claude creator nearly doubles its value from $183B round just three months ago. The startup signed a term sheet with investors including Coatue Management and Singapore's GIC, marking its third mega-round in a year.
Utah becomes first state for AI prescription renewals- Utah partnered with Doctronic to let AI autonomously renew prescriptions for chronic conditions, covering 190 eligible medications with 99.2% accuracy matching physician recommendations. A dozen other states including Texas, Arizona, and Missouri are expected to follow in 2026.
Stanford SleepFM predicts disease from one night's sleep- Stanford researchers developed AI trained on 585,000+ hours of sleep data that predicts dementia (0.85 accuracy), heart attack (0.81), heart failure (0.80), and all-cause mortality (0.84) years before symptoms appear by analyzing physiological signals across brain, heart, and breathing.
ChatGPT web traffic share drops below 70% as Gemini climbs to 20% First significant market share movement in AI chatbot space since ChatGPT's launch signals users diversifying AI tool usage as the assistant market matures beyond single-platform dominance.
ByteDance plans $14 billion NVIDIA chip spend in 2026- One of the largest single-company AI infrastructure investments underscores escalating compute requirements for training and running large-scale AI models and ByteDance's commitment to maintaining competitive AI capabilities for platforms including TikTok.
Caterpillar partners with NVIDIA on construction AI- Heavy machinery giant pilots "Cat AI" assistant in mini excavators and demonstrates digital twin technology at CES. The partnership aims to enhance equipment operation, predictive maintenance, and construction site efficiency through AI.
India extends AI copyright policy deadline to February 6- Government extends public feedback deadline on generative AI copyright working paper by 30 days, signalling deliberate approach to frameworks that could reshape how AI companies operate globally with potential 5-15% royalty payments from AI companies to creators.
Community
The Spotlight Podcast
When Productive Laziness Meets AI: Peter Taylor on the Critical Thinking Crisis
Peter Taylor thought his work on AI was finished. After sounding the alarm in 2022, he stepped back, confident the profession would carry the torch forward. But something nagged at him—a realisation that refused to let go. AI wasn't just another technology shift demanding attention. It was the ultimate expression of productive laziness, the very concept he'd introduced 15 years earlier. That connection pulled him back in.
In this week's episode, Taylor unpacks three forces colliding to reshape project delivery: mind-melting processing power, post-pandemic culture shifts, and conversational AI requiring zero programming skills. But the conversation takes an uncomfortable turn when he raises the question nobody's asking: where will the critical thinking sit when AI handles routine project tasks?
Taylor's answer challenges assumptions. Critical thinking won't disappear—it will migrate away from individual project managers to somewhere else in the organisation. The problem? Most companies haven't figured out where that somewhere is. He illustrates this with examples from Heathrow Terminal 5, where human project managers caught risks no algorithm would have flagged—decisions that seemed logical based on data but would have been disastrous in practice. As junior project managers grow up in AI-augmented environments, will they develop the judgement to recognise when outputs are catastrophically wrong?
The vision that emerges sees project managers becoming "invisible"—not through frantic detail management, but because their unique value lies in thinking no machine can replicate. Taylor's evolved from sceptic to believer, but his conviction comes with caveats about contextual prompting, verification skills, and knowing when to trust versus question AI. Want to know what actually happened at Heathrow Terminal 5? Curious why Taylor specifically speaks early at AI-heavy conferences? He also reveals his personal productivity secret and the one question nobody ever asks him but absolutely should.
Event of the Week
IPTC Summit on AI and Digital Solutions for the Energy Industry
13–14 January 2026 | Conrad Hotel, Dubai, UAE
The IPTC Summit brings together global energy leaders, decision-makers, technologists, and domain experts to examine how artificial intelligence is reshaping upstream operations and the broader energy value chain. Organised by four professional societies (AAPG, EAGE, SEG & SPE), this flagship event serves as the only fully integrated major joint gathering in the Middle East and Asia, fostering collaboration amongst geoscientists and engineers on integrated technical solutions. Under the theme "Transforming Energy Through AI: Innovation, Intelligence, Impact," the summit explores how the energy sector is shifting towards smarter, more efficient operations through AI adoption whilst promoting impactful decarbonisation solutions. The event also hosts the IPTC Excellence Awards, recognising standout technological achievements across the global energy industry. Details here
One more thing
That’s it for today!
Before you go we’d love to know what you thought of today's newsletter to help us improve The Project Flux experience for you.
See you soon,
James, Yoshi and Aaron—Project Flux
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
1