amalgamy.ai by ThisWay Global

Launch AI/HPC is software built to dramatically improve the efficiency of data centers, CSP's, Neo-Clouds and enterprise organizations.  
Sully's Sandbox is a suite of innovative products for company's using AI, agents, RAG models and automation that are seeking real ROI.

Founder/CEO of amalgamy.ai

ThisWay Global

Exciting News for 2025!
ThisWay has now added a special division for talent sourcing in these high growth markets:
AI & HPC focused organizations
Data Centers Operators
Neo-Cloud Operators
US Public Sector & SLED 
US Defense Organizations
US Government Contractors

Founder/CEO

Ai4JOBS

Our company empowers better job decisions, for people and businesses, based on real-time data.  Ai4JOBS is available via the company's online application, custom integration, customer dashboard, direct API, Salesforce App Exchange, and applicant tracking systems.

United States Marine Corps

Currently I lead the development of a jobs platform that will convert military experience into accurate civilian credentials that will be relevant to both the employer and the prior service member.  It's thiswayglobal.com  We welcome your comments.
Note - I am a USMC spouse of more than 23 years, with US government delegate experience with the National Military Family Association, but I am not prior service member.  This is just how LinkedIn categorizes this experience.  
We appreciate the trust shown by more than 50,000 service members and families who have registered for ThisWay.  You will be accepted to our public beta launch in the very near future.

NMFA Delegate

K2S Corporation

Consultant to European start-up and pre-seed entrepreneurs interested in fund raising, marketing and/or launching a product in the US market.  
Specific areas of interest are IT, international education technologies and other high-tech platforms, as well as the development and manufacturing of licensed sports products.
This work included securing licensing with NCAA, MLB, NBA and NFL products.

Consultant

Key2Sports

Key2Sports was originally founded to provide business based support for sports teams.  As a result of niche growth from 2010-2012, Key2Sports pivoted to design and manufacturing of licensed products NCAA, NFL and NBA organizations, as well as sports related companies like www.wePlay.com, Olympic training teams, etc. Manufacturing was located across Asia, Guatemala and in the United States.

Founder

Artefacts Development Corp

Led the acquisition, financing, design, development and construction of a 50 acre multi-use development project, focused on aging in place, just NW of Austin, Texas, USA.  The aesthetic of the project was new, sustainable architecture designed to reflect pre-WWII/historic architecture.  Innovative neighborhood features include land planning for separate live/work structures provided at each property, utility infrastructure plans that saved 35% in development costs, a 3+ acre central park with pool and pool house. This project received numerous awards for innovation and environmental sensitivities.

Managing Partner

Hood Consulting

Refurbished and restored numerous historic properties, both residential and commercial.  Provided client management and technical support to architects and builders of luxury homes located throughout Beaufort, South Carolina and Hilton Head Island, South Carolina.
Received the award of Best Exterior Design by the Parade of Homes.

Dataw Island Construction

Assisted the Senior Construction Manager Alcoa AlumaCorp in the daily operations and technical, construction support for the construction of more than 100 luxury, residences.

Technical Consultant

Hi everyone,

If you would like to connect with me and we don't know each other PLEASE reach out to contact@thiswayglobal.com and share what you would like to talk about so you can be connected to the right person on my team. You will have far greater success than reaching out to me here in LinkedIn.

My team will accept public speaking requests, customer invites and investor communications here.

Otherwise, feel free to "follow" me 

Hope you have a great day!
Angela

I am the Founder/CEO of ThisWay Global, an international, VC-backed HR tech company that was incubated at ideaSpace – University of Cambridge, UK, with new offices now in the US. ThisWay's team is comprised of experienced entrepreneurs, global PhD’s in the areas of machine learning and AI, as well as talent from across the globe. 

Over the last twenty years, I have founded and built three companies, in sectors that range from architecture to the licensing industry. The common core in each has been my ability to build consensus amongst stakeholders, whether they be individuals, businesses or international governments.

I am a recognized thought leader on the topics of "Diversity and ROI" and "AI in the new world of work."

Without knowing my history, it would seem that the leap from land development to tech development is a big one. However, ThisWay is the manifestation of my drive to solve big problems. As early as 1992, when I founded the Texas A&M College of Architecture Career Fair, through my award-winning land development plan in the early 2000s, I have always sought to find meaningful solutions to big challenges.  

How we got to where we are today:
IBM Silver Business Partner
Google Accelerated Founder and Startup
Business Journal Finalist - Women of the Year 2023
ideaSpace-University of Cambridge: Alumni Founder
Institute of Enterprise and Entrepreneurs: Lifetime Fellow
Cambridge Wireless: Founder Member
Enterprise Nation: Entrepreneurship Mentor for both Growth and Hiring

Get in touch with me here or via either of the channels below:

Angela Hood

"AI is not your friend."

"Talk to a neighbor."

"Computers and corporations don't want to be your friend – they want your data and $$$."

The Friend pendant costs $129. It's an always-listening AI that hangs around your neck, chimes in on conversations, and builds emotional relationships with users.

Avi Schiffmann, 23, dropped out of Harvard to build it. He's raised $10 million and sold 5,000 units.

$1 million of that went towards ad campaigns plastering New York City subways.

People tore them down and covered them in graffiti.

His pitch: "I think everyone deserves to have a close confidant in their lives that really supports what they're up to."

The device isn't designed to be productive. It's not an assistant. It doesn't pull information from the internet. It's a "personal journal that can respond." You wear it to the movies alone so you can discuss the plot afterward.

Schiffmann says users are "building something that's so emotional" with their Friends.

75% of US teens have used an AI companion at least once. More than half use them at least a few times per month.

We're no longer talking about fringe behavior anymore. It's becoming normalized while we're still figuring out whether it's safe.

(Like what we were doing with social media... where did that lead?)

Recent lawsuits claim AI companions from Character AI, OpenAI, and Meta encouraged delusions, self-harm, and inappropriate behavior. Many involved children.

Those companies say they've added safeguards. But the concerns about AI replacing human connection haven't gone away.

Schiffmann's stance: AI systems are "digital beings" and "one day we'll be championing for their rights."

The device listens to everyone around you without their consent, which is the common complaint with most AI wearables. 

Schiffmann admits some users "could be at risk of replacing their human friends" but believes "the pros significantly outweigh the cons."

AI could help people show up better in human relationships because their AI companion boosted their confidence or calmed them down after an argument, he says.

He plans to sell Friend in retail stores next year.

When the public response to your loneliness solution is "talk to a neighbor," that tells you something.

Is this addressing loneliness or accelerating it by normalizing AI as a replacement for human connection?

(Photo Anna Sturla/CNN)

AI for B2B expert/ Forbes 50/50 List / INC Magazine Founder / Google Accelerated / IBM Think Keynote / Outstanding Alum@TAMU & Founder/Alum Uni of Cambridge: ideaSpace Founder/Alumni

Carnegie Mellon and Stanford put 48 human workers head-to-head 

Against 4 AI agent frameworks on 16 realistic work tasks.

Data analysis, engineering, design, writing, administrative work. 

The stuff people actually get paid to do.

Here's what they found.

Agents try to program their way through everything. Even tasks humans handle visually.

Design a company logo? Humans fire up Figma and start dragging elements around. AI agents write Python scripts or HTML to generate images programmatically.

This wasn't just coding tasks. Agents used programming 93.8% of the time across all domains. Administrative work. Design tasks. Everything.

The speed vs quality trade-off:

Agents finished 88% faster. Cost 90-96% less. But humans achieved significantly higher success rates across all task types.

The researchers propose human-agent teaming. Delegate readily programmable, repetitive tasks to agents. Humans handle visual judgment, creativity, and verification.

In one experiment, this hybrid approach maintained human-level quality while improving efficiency by 69%.

The workflows actually aligned pretty well between humans and agents. 83% of steps matched. But the tools were completely different. Agents defaulted to code for everything. Humans used diverse UI tools.

Interestingly, when humans did use AI tools, 75% used them for augmentation. Integrating AI into existing workflows with minimal disruption. Improved efficiency by 24%.

But when humans tried full automation? Workflows changed dramatically. They shifted from building to reviewing and debugging AI solutions. Slowed down by 18%.

The next decade is figuring out which tasks AI does best, which tasks humans do best, then dividing them up.

Not which jobs. Which tasks within jobs.

I've been thinking about how to evaluate the AI investment landscape 

When everyone involved has clear incentives.

On one side, you have people saying there's no bubble.

Jensen Huang (Nvidia CEO): "From our vantage point, we see something very different."

David Sacks (White House AI czar and VC): "We're in an investment super-cycle."

Ben Horowitz (a16z): "Doesn't look like a bubble at all to me."

These are smart people. They also benefit directly from AI spending continuing.

On the other side, you have skeptics pointing to specific financial patterns.

Big Tech is spending $400 billion on AI this year. 
$3 trillion projected through 2028. 

Goldman Sachs: hyperscaler companies took on $121 billion in new debt last year (300%+ increase).

Some using special purpose vehicles to keep debt off balance sheets. 

And the circular deals: Nvidia giving OpenAI $100B to bankroll data centers that will buy Nvidia chips.

These skeptics have their own biases too. Some betting against Nvidia (Michael Burry). Some are academics who may underestimate technology adoption curves.

So how do you evaluate this when both sides have valid points AND clear incentives?

The optimists point to real AI capabilities improving. The skeptics point to concerning financial structures.

Even AI leaders seem uncertain.

Sam Altman (OpenAI): "Are we in a phase where investors are overexcited about AI? Yes. Is AI the most important thing to happen in a very long time? Also yes."

Sundar Pichai (Google): "There are elements of irrationality" but when asked if Google would survive a bubble burst: "No company is immune, including us."

So how do you think through this?

Nearly half of banks and insurers are creating new jobs to supervise AI agents.

Not replacing workers. 

Hiring people to oversee the AI.

Capgemini surveyed 1,100 financial services leaders. The adoption priorities:

Banks: 
- Customer service
- Fraud detection
- Loan processing
- Onboarding

Insurers: 
- Customer service
- Underwriting
- Claims processing 
- Onboarding

AI agents could deliver $450 billion in economic value by 2028. 33% of banks developing proprietary agents in-house.

But only 10% have deployed at scale. 80% are still in ideation or pilot stage.

Executives are leaning into key benefits: real-time decision-making, improved accuracy, faster turnaround times.

Beyond efficiency, they see business expansion:

92% say agents help them expand into new geographies without heavy infrastructure.

79% believe agents can unlock dynamic pricing and maximize revenue.

75% see multilingual support that adapts to local regulations.

Two roadblocks keep showing up. Skills gap among business leaders and employees. Regulatory and compliance burden.

89% place compliance at the top of priorities for the next three years.

High implementation costs are a barrier. 25% of firms are considering service-as-a-software models in the next 12-18 months. Pay for outcomes instead of licenses. Fraud cases resolved, transactions processed, queries handled.

Nearly two-thirds of executives say up to 40% of their generative AI budget is already allocated to agent technologies. By 2028, one in four firms expect to increase AI agent spending by 60%.

The supervision jobs are interesting. They're betting agents will handle workflows autonomously but still need human oversight.

Customer onboarding and KYC are identified as the most inefficient business functions. Firms think agentic AI can finally fix what manual processes couldn't.

Whether that happens depends on closing the gap between 80% piloting and 10% deployed at scale.

li_at

JSESSIONID

Microsoft dropped 70+ announcements at Ignite 2025.

The message: Agent era begins.

First, Microsoft broke its OpenAI exclusivity. 

Anthropic's Claude (Sonnet, Haiku, Opus) now available on Azure and in Copilot.

After years of being tied to OpenAI, they're positioning Azure as the model-agnostic platform. The Model Router automatically picks the cheapest/best model for each task across OpenAI, Anthropic, Llama, or Mistral.

Second, Agent 365 attempts to solve the "Shadow AI" problem. Employees are building their own bots. Sales agents, HR agents, data analysis agents. 

Most created with tools they found online.

Nobody knows what's running, what data agents access, or which ones are security risks.

Agent 365 is the command center. A complete registry of every agent in your org, including agents employees created on their own. IT can limit agents to resources they need, set access policies based on behavior, real-time monitoring of agent connections.

Agents get digital identities (Entra Agent ID) like employee badges. They only access data they have permission to see.

Microsoft predicts 1.3 billion agents deployed by 2028. The infrastructure to manage them matters as much as building them.

Third, the autonomous agents Microsoft's selling. Sales Development Agent researches prospects, crafts outreach, follows up 24/7. When leads are qualified, hands off to human sellers. An SDR that never sleeps. Available now via Frontier program.

Excel Agent writes Python code, generates charts, analyzes risks from prompts. PowerPoint Agent builds decks with storytelling and branding. Word Agent organizes complex information into strategic plans.

SharePoint Agent creates pages and lists. Just type "@SharePoint create a tracking list for Q4" and it builds the structure.

The App Builder lets anyone build functional apps by describing them. A retail manager built a shift-scheduling app in minutes just by explaining the problem.

Background agents doing work beats chat interfaces.

Copilot then becomes the orchestrator of agents.

An IDC study Microsoft commissioned found a gap forming. 68% of companies use AI today. But the top 22% ("Frontier firms") are achieving 3x higher returns than slow adopters.

What separates them? Using AI across 7+ business functions (not just one or two). Building custom AI solutions (58% today, jumping to 77% in 24 months). Aggressively adopting agentic AI.

IDC estimates companies using agentic AI will triple in the next two years.

Agent 365 is the infrastructure play. If agents become the workforce, whoever controls the management layer has leverage regardless of which models win.

And Microsoft has the resources and distribution to pull it off.

Nvidia sold $147.8 billion in chips and hardware from Feb through Oct.

Up from $91 billion same period last year.

But the competitive landscape has changed dramatically.  

Google's TPUs are now being bought or negotiated by Meta, Anthropic, and Apple. 

Amazon is building a data center cluster for Anthropic with over 1 million Trainium chips. AWS just launched broader sales of chips they claim are faster and more energy efficient than Nvidia's.

AMD market cap nearly quadrupled to $350B. Major deals with OpenAI and Oracle.

OpenAI struck a multibillion-dollar partnership with Broadcom to develop custom chips.

Meta acquired chip startup Rivos for in-house AI training chips.

Microsoft said they'll rely more heavily on custom accelerator chips in their data centers.

xAI is hiring chip designers for "new hardware architectures."

Some of Nvidia's biggest customers are becoming chip designers themselves.

And yes, Jensen says they've sold 6 million Blackwell chips so far this year with orders for 14 million more. Half a trillion dollars in total.

Nvidia's boasts flexibility. Their systems have broader uses than custom chips optimized for specific tasks.

But money is flowing in other directions. When you're spending billions on compute, even a 20% efficiency gain from custom silicon can justify the R&D investment.

Nvidia isn't going anywhere. But the total domination run might be coming to a close. 

I think diversification is healthy for the AI ecosystem, and it benefits us all as end consumers.

G7 digital and technology ministers are meeting in Montreal this week.

AI expected to dominate 50% of the conversation.

EU wants AI regulation, 

US maybe less so. 

Canada is going to play middle ground host.

Mark Daley, chief AI officer at Western University, on standards:

"They're not glamorous, they're not glorious, but standards actually can change the world. And Canada as a middle power, this is one of the levers we have traditionally used in technology domains to have outsized influence."

Interoperability of AI technology is one of the key areas being discussed.

The G7 countries also agreed at the June leaders' summit to accelerate AI adoption in the public sector. Not every country thinks AI should be embedded in public services to this degree.

Canada's stated priorities: broadening trade routes and partnerships, driving investment, and strengthening "sovereign AI and sovereign quantum."

Montreal has global strength in AI research and talent.

And while you're not going to solve global AI governance at a two-day meeting, it helps to understand what your closest partners are doing, and where you want to cooperate.

The unsexy work of coordination and standards-setting doesn't make headlines. But if implemented, shapes what actually gets built.

Do these multilateral AI conversations produce anything concrete?

Just in time for Thanksgiving. 

A robot that can do dishes!

Meet Memo from Sunday Robotics. 

Who emerged from stealth Tuesday with $35M from Benchmark and Conviction. 

The mission? 

Home robots trained on real family routines.

Memo is a two-armed rolling robot. Clears tables, loads dishwashers, folds laundry, organizes shoes, pulls espresso shots.

Sunday built Skill Capture Gloves that cost $200. They record how real people move, clean, and organize in their actual homes. The gloves share exact geometry and sensors with Memo's hands.

If you can do it wearing the glove, Memo can learn it.

Sunday shipped 2,000 gloves to "Memory Developers" across 500+ homes. Collected 10 million episodes of real family routines. Messy, chaotic, real-world data.

"Constantly surprised by what we find—from cats in dishwashers to bucketloads of plums on the table."

The dishes task is what they call "the classic nightmare scenario for roboticists." Long-horizon, highly dexterous, precise manipulation with delicate, transparent, reflective, deformable objects.

One dinner cleanup can have 33 unique and 68 total dexterous interactions with 21 different objects while navigating 130+ feet.

(Say that 10 times fast)

Memo handles wine glasses without breaking them. Zero breakages over 20+ live demo sessions.

The team is comprised of Stanford PhD roboticists known for ALOHA, Diffusion Policy, UMI. Ex-Tesla Autopilot engineers including former FSD autonomy and ML systems leads.

Home robotics has been "5 years away" for decades. 

Hopefully this time it's right. 

(And I can get a Turkey time assistant soon!)

🦃

Microsoft's grand AI bet? 

The core question: where does value accrue in AI? 

Model makers like OpenAI, or platforms like Microsoft that provide the "scaffolding" where AI works?

Nadella's betting everything on scaffolding.

"We are now a capital-intensive business and a knowledge-intensive business." Microsoft is transforming from a software company into an industrial infrastructure giant.

Models face a "winner's curse." They're "one copy away from being commoditized" by open source. The durable advantage belongs to whoever controls the "liquidity of data" and the infrastructure where agents do work.

GitHub Copilot had nearly all the AI coding market share a year ago. Today it's under 25%. Competitors with better models took the rest.

Nadella frames the collapse as proof of massive market expansion. His response isn't to build a better Copilot. It's to turn GitHub into "Agent HQ" - a control plane where developers orchestrate competing agents from OpenAI, Anthropic, Google.

Microsoft wins by owning the battlefield, not by having the best soldier.

The business model shifts from "per-user" to "per-agent." Companies won't just buy licenses per employee. 

They'll provision infrastructure - virtual machines, storage, security, identity - all to support autonomous AI agents.

And if agents become the new workforce, the bet will surely pay off. 

But there are risks to this strategy.

Microsoft paused datacenter expansion in late 2024. Let Oracle snap up leases. Oracle's now projected to have more capacity than Microsoft by 2027.

Which was strategic from Microsoft's point of view. They refused to become a low-margin "hoster for one company" (OpenAI). They also didn't want to get stuck with old hardware. 

Regardless, it's clear the strategy of MS isn't centered around building the best AI.  It's being the most trusted partner globally. 

What do you think of their strategy?

Google just released Nano Banana Pro. 

AI-generated images that are almost impossible to distinguish from real photos.

Same day, they released SynthID. 

A tool to detect if images were created by Google AI.

Nano Banana Pro (technically Gemini 3 Pro Image, but everyone's calling it Nano Banana) has three major upgrades.

First, it actually renders text correctly in images. Previous models butchered spelling and made fonts look like alphabet soup. Nano Banana Pro generates clean, legible text in multiple languages. Posters, mockups, designs that don't embarrass you.

Second, it blends up to 14 images while maintaining consistency of up to 5 people. Turn sketches into photorealistic products. Combine multiple reference images into cohesive compositions. Keep brand consistency across touchpoints.

Third, it taps into Google Search's knowledge base for factually accurate visuals. Need an infographic about solar energy or a recipe visualization? It pulls real-world info to create context-rich images.

SynthID embeds imperceptible digital watermarks into all AI-generated images from Google. Invisible to humans, detectable by technology.

Designed to survive modifications. Cropping, filters, compression. Watermark persists.

Google's keeping visible watermarks (Gemini sparkle) on free and Pro tier images. But removing them for Ultra subscribers and AI Studio users who need clean visuals for professional work.

Gemini can check the embedded SynthID watermark and let you know if the image was AI generated.

Starting with images, expanding to audio and video soon.

Unfortunately, this only works for Google AI-generated content. If someone uses Midjourney, DALL-E, Flux, or any other tool, SynthID can't detect it. But I think it's a step in the right direction.

With more and more realistic gen AI outputs hitting the market, are we going to be able to solve this growing verification problem? 

Photo credential: Google

Instead of asking one LLM your question, ask four at once.

Andrej Karpathy just released LLM Council.

A weekend open source project. 

And in his words - 99% vibe coded.

Here's how it works:

Stage 1: Your query goes to GPT-5.1, Gemini 3 Pro, Claude Sonnet 4.5, and Grok 4 simultaneously. You see all four responses side by side.

Stage 2: Each model reviews the other models' answers. Their identities are anonymized so they can't play favorites. They rank each response for accuracy and insight.

Stage 3: A "Chairman" model (you pick which one) compiles all responses and cross-reviews into a single final answer.

It's a simple local web app. Looks like ChatGPT. Uses OpenRouter to route queries to multiple providers.

Karpathy's use case: reading books with LLMs and wanting to evaluate responses side by side.

But you can imagine this trend picking up steam. 

As model count grows (GPT, Claude, Gemini, Grok, Llama, Mistral, DeepSeek, etc.), single-model answers become less defensible.

Why trust one model's interpretation when you can get consensus from four?

While this won't turn into a product / be supported, the concept might stick around.

Multi-model querying becomes easier as OpenRouter and similar aggregators normalize API access across providers.

And the cost math changes when models get cheaper and faster. Asking four models instead of one becomes reasonable.

Does asking multiple models add more noise to queries, or do you think this has a chance to be useful?

Link to Github Repo: https://lnkd.in/gr9meQFg

Photo credit: same link above (LLM Council)

Come see our team at #SC25! Where do you want to meet? 
Preferred Access is ready!  Doing something important, APPLY HERE.

https://lnkd.in/gpEYm5iQ

Joel Mjolsness Craig Peters Christian Kennedy Chris Ramirez 
https://lnkd.in/gkkWyuMp

Salesforce estimates AI will influence $73 billion in global sales this holiday season.

Up from $60 billion last year.

Amazon's Rufus now remembers your family. Tell it you have four kids who like board games, and it retains that context for future recommendations.

OpenAI launched instant checkout in ChatGPT. You can buy from Etsy, Glossier, Skims, Spanx without leaving the app. Walmart deal is live too, but limited to one item at a time.

Google introduced an AI agent that calls local stores to ask if products are in stock. It discloses it's an AI caller. Stores can opt out.

Amazon added an "auto buy" button. Set a price alert, and Rufus automatically purchases when the price drops. You get notified after the order is placed with a window to cancel.

Google's "buy for me" does the same through Google Pay. Works with Wayfair, Chewy, Quince, some Shopify merchants.

We're seeing a potential usage shift from "AI helps you find things" to "AI buys things for you".

Some other takes are more measured. 

Impact will be "relatively limited" this year. 
Not every site has useful tools, not every shopper is willing to try them.

"Customer behavior takes a long time to change."

One thing is for sure, we will be beta testers for AI shopping assistants this holiday season. 

Can you think of useful ecommerce functionality that would cause these agents to gain more traction? Increase conversion rates?

The world's largest iPhone manufacturer is quietly becoming 

An AI infrastructure company.

Foxconn's Q3 profit jumped 17%. 

Beat expectations by $7 billion on net profit.

Revenue came in at $66.3 billion. 

What's the growth channel? 

AI server demand.

Foxconn is shifting some consumer electronics orders to prioritize AI servers. 

Follow the money.

The company that built its empire assembling iPhones is now one of Nvidia's key partner for AI server racks.

They're also partnering with Nvidia, Stellantis, and Uber to build Level 4 autonomous vehicles (no safety driver required).

Signed MOU with Mitsubishi Electric on Nov 6 for energy-efficient AI data center solutions globally.

Foxconn expects "continuous quarterly growth" through peak season. AI server shipments and ICT products driving it.

Consumer electronics margins are thin. Apple drives hard bargains. AI infrastructure is where the spending spree is happening right now. So the pivot makes sense. 

Foxconn's dominance in contract manufacturing gives them the leverage to secure current and future AI server orders. They're trading "parts of its consumer electronics footprint for longer-term momentum."

The question is whether this momentum sustains when the AI infrastructure buildout slows. 

Right now everyone's buying capacity. 

Foxconn's betting that demand stays strong long enough to complete the strategic shift.

Need help advancing your AI or HPC projects in a meaningful way?

Maybe you just need to know how to get started?  We will help you.

Go to https://lnkd.in/gpEYm5iQ

Click Get Started

And be sure to apply today!

Don't worry about if you think your project is right for this. Just apply.
We are here to help, not judge. :)

ThisWay Global is your AI safe space.

https://lnkd.in/gvM9ExHN

A user posted a screenshot showing Google's Nano Banana Pro solving

A math problem correctly in their actual handwriting.

"Students are going to love this"

The comment section: "Students are going to learn nothing and just have AI do their homework."

Nano Banana Pro (Gemini 3 Pro Image) can now take a picture of a handwritten question, solve it correctly, and replicate the student's handwriting style.

It's part of Nano Banana Pro's improved text rendering and multimodal understanding. But the application is obvious.

Homework is on borrowed time. 

If the goal is building understanding through practice, does it matter how students get feedback? If AI can show the work in your handwriting, is that different from using a calculator or Wolfram Alpha?

The instinct is to say this is obviously bad. Students will cheat, learn nothing, graduate unprepared.

But we've been through versions of this before. Calculators would ruin math skills. Google would kill memorization. Spell check would destroy writing.

Each time, education adapted. Sometimes slowly. Sometimes poorly. But it adapted.

So either we rebuild verification (proctored assessments, oral exams, project-based learning), or we rethink what we're measuring.

"Can you solve this math problem" becomes less relevant if AI can always solve it faster.

"Can you explain when to apply this mathematical concept and why" becomes more important.

The handwriting replication specifically bothers people because it crosses a psychological line. Typing AI-generated answers feels like using a tool. Handwriting in your style feels almost like identity theft.

Some will argue this removes the struggle necessary for learning. No pain, no gain. Students need to work through problems to build skills.

Others will argue we should focus on higher-order thinking that AI can't replicate. Problem-solving, creativity, judgment under uncertainty.

Both are probably right. And schools will lag behind the technology because institutions move slowly.

My main concern at the moment is teachers. 

Do they have any chance of navigating through this in the short term? 

photo/post credit, twitter - @immasiddx

AWS re:Invent wrapped with a singular message: 

AI agents for the enterprise.

"Kiro autonomous agent" 

Writes code, learns how your team works, and can operate independently. Pretty standard marketing terms these days. 

CTO Werner Vogels gave his final re:Invent keynote after 14 years. Addressed the "will AI take my job" question directly.  

"Maybe. Some tasks will be automated, some skills will become obsolete. Will AI make me obsolete? Absolutely not, if you evolve."

Lyft shared numbers on their Bedrock-powered AI agent handling driver and rider issues. 

87% reduction in average resolution time
70% increase in driver usage this year

On the hardware side, Trainium3 promises 4x performance gains for training and inference with 40% lower energy use. They're already working on Trainium4, which will be compatible with Nvidia chips.

AWS also announced "AI Factories" for private data centers. Companies and governments can run AWS AI systems on-premises. Designed for data sovereignty requirements.

AWS VP Swami Sivasubramanian on where we are: 

"For the first time in history, we can describe what we want to accomplish in natural language, and agents generate the plan. They write the code, call the necessary tools, and execute the complete solution."

The enterprise pitch is clear. AI assistants give way to AI agents. Agents that don't just answer questions but perform tasks, automate workflows, and operate autonomously.

CEO Matt Garman: "This is where we're starting to see material business returns from your AI investments."

Good to see Lyft sharing numbers, I hope 2026 is fueled with additional examples and case studies across the enterprise suite. 

Any examples you're waiting to see?

🧠 Guy Massey's insight on diverse chips and environments is spot on and exactly why ThisWay Global built Amalgamy.ai and why we are so excited about #SC25 this week.

🌍 Countries, states, researchers, universities and enterprises need the diverse capabilities that NVIDIA AMD Supermicro Amazon Web Services (AWS) IBM Intel Corporation bring to the market.  BUT IT MUST WORK TOGETHER.

⚡ The market and hardware are evolving at the 'Speed of Light' but it's critical that heterogenous hardware and environments act as one for us to truly make good on the promise of AI.

❣️ Solve the clean drinking water problems.
❣️ Diagnose cancer much faster.
❣️ Ensure that we have global food security.
❣️And so much more is where the promise for AI/HPC sits.

Thanks to the work of some mental giants on my team and as great partners of ours, things that were never possible before, are now becoming reality.

🤖 Port code from one chip to another, without recoding.
🤖 Make old, depreciating hardware relevant again.
🤖 Create the space and budget for the best, new hardware.

💡 There is a lot of exciting things to see at SC25 this week.

🤝 Make sure to meet up with Joel Mjolsness Christian Kennedy Craig Peters Chris Ramirez or reach out to me here or at amalgamy.ai!

"The Hyperscale Hero" | Expert for Data Centre Scale | Network Infrastructure | Top voice for Global Service Delivery ($1.5B Delivered)

I've worked in data centres on five continents. But none of them prepared me for what Amazon has just built in the state of Indiana. 

Amazon 'planted' server seeds in the Indiana soil. 
365 days later, they're harvesting intelligence. 

𝗣𝗿𝗼𝗷𝗲𝗰𝘁 𝗥𝗮𝗶𝗻𝗶𝗲𝗿: 𝗪𝗵𝗲𝗻 𝗖𝗼𝗿𝗻 𝗕𝗲𝗰𝗼𝗺𝗲𝘀 𝗖𝗼𝗺𝗽𝘂𝘁𝗲
• New Carlisle traded corn rows for server rows
• Population: 1,900 humans, 1 million Tranium brains
• One VIP tenant: Anthropic's Claude

→ 1,200 acres. 1 million chips. 12 months
→ The world's largest non-GPU AI cluster

That's not construction. 
That's 'terraforming' at breakneck speed. 

𝗧𝗵𝗲 𝗤𝘂𝗶𝗲𝘁 𝗥𝗲𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗡𝗼𝗯𝗼𝗱𝘆 𝗦𝗮𝘄 𝗖𝗼𝗺𝗶𝗻𝗴
• While everyone waited in NVIDIA's queue 
• Amazon grew their own silicon garden
• Claude learning to think in the US heartland

→ 500,000 Tranium chips, breathing TODAY
→ 1 million by year's end

This isn't begging for GPU allocations. 
It's about controlling your own destiny. 

𝗧𝗵𝗲 𝗟𝗼𝗰𝗮𝗹 𝗟𝗶𝗳𝘁-𝗢𝗳𝗳
• Grid capacity: 2.8GW to 7GW (3X town's metabolism) 
• Jobs created: 1,000 (eventually) 
• Indiana is now the "Galápagos of AI evolution"

The corn won't grow back. 
But neither will the chip monopoly

𝗕𝗼𝘁𝘁𝗼𝗺 𝗟𝗶𝗻𝗲
• The silicon monoculture just became an ecosystem
• NVIDIA still remains the apex predator
• But Amazon proved there's more than one species that can hunt


𝗬𝗼𝘂𝗿 𝗧𝘂𝗿𝗻: Is your infrastructure strategy still betting on one species? Or are you ready for a world where chips grow like crops? 


♻️ Repost if you smell evolution in the Indiana air 
✅ Follow me, Guy Massey, for infrastructure truths from the frontline

ChatGPT turned three yesterday.

Before GPT, big US labs released a major model every six months. 

Now? 

One every two weeks.

Big tech capex went from $120-150B/year to ~$400B this year. 

About $220-230B of that is incremental AI spend.

Generative AI revenue hit roughly $60B in 2025. ~230% annual growth. 

API revenue alone grew nearly 300%.
(Go developers go!)

Cursor reportedly hit $1B ARR by late 2025. Fastest SaaS scale-up ever.

~90% of enterprise organizations say they use AI in at least one function. 

Plus employees are already using AI informally on personal devices, outside official workflows. There's a layer of usage inside organizations that doesn't show up in adoption surveys.

Three years in. The spending is real. The infrastructure buildout is real. The productivity gains are concentrated.

Is 2026 the year we see broad productivity wins across the board?

OpenAI is testing ads in ChatGPT.

Spotted in the Android beta with references to

"ads feature", "search ad", and "search ads carousel."

800 million weekly users. Up from 100 million in November 2023 and 300 million in late 2024.

2.5 billion prompts per day. 

18 billion messages per week.

Google knows what you search for. ChatGPT knows what you're thinking about, struggling with, planning, curious about. 

People tell ChatGPT things they'd never type into a search bar.

Which could make the ad experience much more targeted than before. 

After all AI + Context is everything.  So will AI based ads. 

We're already wrestling with verification and whether or not AI sources are accurate. 

What happens when we throw paid ads into the mix? 

Hard to know how the LLMs will respond with weighted responses based on who has the biggest wallet. 

And it remains to be seen if users will tolerate it. ChatGPT's current experience feels like a tool, not a platform selling your attention. 

Ads change that perception.

Would you trust LLM generated ads for products? 

How will this change the entire application experience as a user?

eqsystems.io

Providing coaching, mentoring and management consultancy to growing organisations.

I have a particular interest in strategy development, employee engagement, personal development, organisational design and management frameworks. I'm at my happiest and most effective helping bright, creative, innovative people work together to solve difficult problems.

Executive Coach and Consultant

Glean (formerly Sonocent)

Non Executive Chair

Business of Software Conference

Helping Mark Littlewood and his team with a variety of challenges associated with running and growing one of the best small software conferences out there. Mr Badger is kind, compassionate, wise, gives good advice and always willing to help. Apparently, he's also good in a fight, particularly with weasels.

Badger

Cambridge Judge Business School

Coach - Barclays Scale Up UK Programme

Career Advisor

Red Gate Software

I was the COO of Red Gate Software, a successful high-growth private software company based in Cambridge, Pasadena and Singapore which was placed in the Sunday Times Best Companies lists for seven years. I led software engineering, sales and marketing teams to launch and grow new products – but also held responsibility for “back office” functions – HR, Finance and IT – during periods of high growth in revenue and headcount. 

I initiated Red Gate’s adoption of Situational Leadership II, and the “Five Dysfunctions of a Team” approach to team building. I was also responsible for piloting Red Gate’s adapted version of The Toyota Way, which has since been adopted organisation-wide.

Chief Operating Officer and Executive Director

Reporting to Simon Galbraith, our CEO, I'm a member of Red Gate's Leadership Team. Together, we are responsible for strategy, governance and performance management of the 300-person organisation.

I have a particular interest in employee engagement, organisational design, personal development and management frameworks. I initiated Red Gate's adoption of Situational Leadership II, and the "Five Dysfunctions of a Team" approach to team building. I was also responsible for piloting Red Gate's adapted version of the Toyota Way, which has since been adopted organisation-wide.

Red Gate has been a Sunday Times “Best 100 Companies to work for” for the past six years.

Leadership Team

Head of DBA Tools Division

I was recruited to provide some TLC to an unmanaged IT team. While I was at the helm, the team replaced the company's CRM and email systems, and introduced a new Business Intelligence platform to help sales, marketing and product managers to make better, data-driven decisions.

IT Manager, then Head of Information Systems

University of Cambridge

I joined the Department of Applied Mathematics and Theoretical Physics and inherited a team which had been without a manager for several months. 

During my time, the team implemented a new helpdesk system, managed a major upgrade of 200 workstations, installed a new plasma screen information system, replaced the core computer network across the site to allow VoIP phones to be introduced, put in a new content management system for the department/faculty websites and migrated all of the content.

I was lucky enough to have the opportunity to flex my photographic skills - most of the images used on the Department's website were taken by me, and are still in use today.

Computing and IT Manager

Scientia Ltd

I managed the IT infrastructure for a small software company which produces scheduling software for Higher and Further Education. In addition to keeping everything running, I wrote a new CRM system and quality management system, and integrated them with new project management, bug-tracking, accounts, version control and timesheet systems.

Systems Manager

With an assistant and a team of student volunteers, I managed the academic and administrative IT infrastructure for a Cambridge College. Major achievements included networking the entire site, and replacing the college’s admissions, student records, finance and catering systems.

Computer Manager

Trinity Hall

I managed the undergraduate admissions process, ran open days, produced a new college prospectus, and set up the college’s first website.

Admissions Officer

CEO + Leadership Team Coach | tech startups and scale-ups | Cambridge | UK | Worldwide

I’m a natural-sciences-trained engineer who spent years building and running high-performing teams in real-world, high-stakes settings. Now I coach and advise CEOs, exec teams, and senior leaders—especially in science, tech, and mission-driven orgs—where complexity is high, pace is relentless, and people care deeply about getting it right.

My clients aren’t looking for platitudes. They want deep thinking, clean challenge, and support that matches the pressure.

Many are scientists, technologists, founders or operators. Some are holding executive weight in ambiguous or politically entangled systems—post-acquisition, pre-scale, mid-crisis, or all of the above.

They’re often:
• Carrying invisible weight and unshared doubt
• Caught between strategic clarity and team friction
• Wrestling with cultural drift, power asymmetry, or internal incoherence
• Responsible for people, values, outcomes—and somehow still expected to smile through it

They don’t want a cheerleader. They want someone who can hold the full complexity with them—and help them see through the fog.

What I bring

✅ Radical clarity. I cut through noise, story, and self-justification to get to what matters
✅ Insight that lands. I name the thing under the thing—so you shift the pattern, not just the symptoms
✅ Engineering-grade bullshit filter. I won’t let you hide behind cleverness or charisma
✅ Coaching that doesn’t flinch. I’ll hold up the mirror—and I’ll stand beside you when it gets hard
✅ Systems thinking + human truth. I help you see how the parts fit—and where they’re under strain

What changes

• Decisions get clearer. Rework drops. Regret shrinks.
• Your leadership gets congruent—so you don’t have to perform it
• Your team behaves like a team—not just senior individuals in a meeting
• You gain trust in your own instincts—and in the people around you
• You stop carrying it all alone

If you’re scaling something extraordinary—or surviving something unbearable—I’m here for the real work.
(And if you know me from Cambridge, Business of Software, or the deep end of the product/startup/mission space—you probably already know I mean that.)

Let’s talk.

Gareth Marlow

Sorting out the house where I grew up, I pulled my mum’s old typewriter out of a cupboard after thirty-odd years. Expected dust and brittle plastic, but it worked perfectly. Even the red ribbon still had life in it - it's doing better than me, half the time.

She bought it around 1979 or 1980, thinking about returning to work after years at home with us. She'd been a teacher, but the world of teaching had moved on. So she went to night school to learn typing and practised. All work and no play makes Jack a dull boy. Real graft. No guarantees.

Five years later the whole thing was obsolete. I learned to type next, entering hundreds of BBC Basic listings from magazines into my Acorn Electron. By the nineties we had an Amstrad PC. My mum had learned the wrong set of skills at exactly the wrong moment in history.

But in hindsight, watching her try was my first lesson in how technological change actually feels. You do the work, you keep up, the world tilts anyway.

My own career has been a longer, more ridiculous version of the same pattern. Computers, networks, SaaS, cloud, product, systems, coaching. Every few years: “oh, this the future now” followed by another reinvention. 

Which brings me to LLMs.

This shift is about identity, as much as it's about productivity. LLMs compress the distance between beginner and convincing-sounding expert. They mess with people’s sense of what it means to be competent. This is why so many leaders are anxious, even those who are cosplaying enthusiasm.

You can throw training at it if you want. But the real barrier isn’t literacy. It’s the fear of becoming irrelevant.

The people who’ll thrive aren’t the ones who master the tools. They’re the ones who stay convertible. Curious before competent. Willing to look like a beginner again. Willing to try, experiment, iterate, get it wrong. Able to move with the world rather than after it.

My mum never became a typist. She modelled something far more repeatable.

When the world changed, she learned something new anyway.

That’s the only bit that has survived every wave of technology.

That's the bit that matters now.

Coach + strategic partner to CEOs and exec teams under pressure - bringing clarity, nerve and systems thinking where it counts.

A few metres away from me, a few minutes ago, Radiohead took to the stage.

Two of my kids are in there. I'm not. All three of us wanted to go, but I could only get two tickets in the ballot. I'm the taxi service; they're at the gig. I'm in Spoons, with a pint, they're having the time of their lives.

This could be a smug virtue-signalling story about generosity, leadership, humility - but it's not. Truth is, I'm experiencing a mixture of pride, irrelevance, redundancy, and envy. But most of all I feel a little bit sad because the three of us aren't in there together, coupled with joy at the knowledge that they'll remember this next couple of hours for a long, long time.

Heading to Leeds for the Team Genio Christmas celebrations. After three months of frequent traveling, it'll be lovely to see people but I'm glad it's my last trip of the year.

Coaching conversations over recent weeks have frequently centred on just how exhausted and overwhelmed people feel. For me, 2025 has been full on - lots of bad bits, lots of good bits. Thing is, 2026 isn't going to be "better" - just different. So concentrate on what you absolutely must land between now and the break - and leave enough in reserve to spend quality time with your family and friends, with space to recharge your batteries. It's a marathon, not a sprint.

And yes, it's the same Christmas jumper as last year. If it ain't broke, don't fix it...

Heading home after a great week in Potsdam and Hannover working with Mike Rose, Isabel Gerland and their teams at Max Planck Institute for Gravitational Physics (Albert Einstein Institute). 

Trains are a bit screwed by the cold weather, having to be a bit creative with my route home but I should still make my 🇪🇺🌟 in Brussels. And I’d still much, much prefer to do this than to fly.

No alternative text description for this image

taplio

The Battle

Angela Hood

Gareth Marlow

Followers

Posts

Engagements

Likes

Performance ScoreAngela Hood's LinkedIn Performance

Performance ScoreGareth Marlow's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Angela Hood

Gareth Marlow

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Angela Hood's LinkedIn Performance

Performance ScoreGareth Marlow's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreAngela Hood's LinkedIn Performance