We're in a strange phase with AI at work.
Most leaders say it's already making a significant impact. Most employees are using it regularly. And yet, when you look at what has actually changed inside organisations, the results are far more modest than the hype suggests.
That gap between ambition and reality is the theme I keep seeing.
This week's reads and reflections touched a few different corners of that story: how companies move beyond pilots, why support still feels like a black hole, how quality assurance is being automated at scale, and why the AI market itself is entering a pressure-cooker phase. And while this post is mostly about workplaces, the same pattern applies to AI for schools too: adoption can move fast, but real transformation only happens when workflows, culture, and capability change together.
Most organisations are still early even if they think they are ahead
A new Google Workspace study on AI in the workplace put a number on what many of us already sense. Only about 3% of organisations in the research are "highly transformed" with AI.
Even though most leaders say AI is already having a significant impact.
A few themes stood out:
- Employees are using AI every day, but many do not feel prepared, empowered, or included in the strategy
- Time savings are common, but the real gains show up when AI boosts innovation, creativity, and new products
- Highly transformed organisations report stronger ROI, better collaboration, higher employee satisfaction, and a clearer edge in the market
The report also outlines recurring steps organisations take to move from pilots to real transformation:
- Build an "always on" AI strategy and roadmap that is transparent across the company
- Make AI part of culture and everyday work, not just a tech project
- Start with quick wins that cut across roles and departments
- Democratise advocacy so champions exist beyond the AI or IT team
This matches what I see in practice. Adoption spreads faster than capability. People use tools, but the organisation hasn't redesigned the work around them yet. Strategy exists in slides, not in the weekly rhythm of how things get delivered.
Read more: Beyond AI optimism - Google Workspace study
The DeepMind story is a reminder that breakthroughs are built on boring discipline
I watched The Thinking Game on the weekend. It traces the story of DeepMind and the broader arc of AI development, from early work on games through AlphaGo and into the scientific impact of AlphaFold.
It's accessible, but it doesn't dumb it down. It blends personal biography, technical explanation, and ethical reflection. It follows Demis Hassabis and his collaborators as they move from stealth startup to acquisition and global attention.
A few points the film highlights:
- Games became controlled testbeds for reinforcement learning
- AlphaGo and AlphaZero challenged long-held assumptions about what machines can learn
- AlphaFold shifted AI from beating games to solving a decades-old scientific problem
One of the best parts is hearing directly from the researchers and engineers. Their mix of excitement, doubt, and pressure is a good antidote to today's "AGI vs doom" noise. It's a reminder that real progress is usually a messy mix of deep thinking, iteration, and uncomfortable uncertainty.
youtube.com/watch?v=d95J8yzvjbQ&t
Service desks still feel like a black hole and that's about to change
I've been reflecting on how service desks fit into our work, and how I avoid them until I absolutely have to log something.
Most people I talk to feel the same. Submitting a ticket often feels slow, impersonal, and a bit like throwing your problem into a system that doesn't talk back. So we send an email instead. Or a Teams message. Or we ask someone we know because it's faster and more human.
Teams on the other side feel the pain too. They need systems to keep work organised, but they know users don't love the process. When people avoid the system, visibility drops and things slip through cracks.
What's changing now is expectations. We want support to feel like a conversation, not a transaction. And with AI improving, more companies are making support easier by hiding the ticket behind:
- email and chat intake
- automated triage
- summarisation and routing
- response drafting and knowledge retrieval
The goal is clear. Keep the structure, remove the friction. Let people ask for help naturally and let the system handle the boring parts in the background.
Big changes are coming in this space, and honestly, it's overdue.
Quality assurance is quietly one of the biggest AI wins in enterprise
One of the most practical examples I saw this week was Cochlear expanding its use of AI inside its Amazon Connect contact centre environment.
A major shift has been automated evaluation of agents and calls. They've moved from about five manual scorecards per agent per month to thousands of automated assessments. That depends on accurate transcription plus a mix of rule-based checks and generative AI reasoning.
The benefits are straightforward:
- wider visibility of quality and compliance trends
- more targeted training based on consistent scoring
- the ability to assess every call instead of a tiny sample
They're also moving toward more data-driven workforce management using historic data for forecasting, capacity planning, and scheduling. And they're exploring early use cases for AI agents that can answer voice calls, with encouraging pilot results.
This is the kind of AI impact I trust most. Not hype. Just better coverage, better feedback loops, and fewer blind spots.
Read more: Cochlear plugs AI into its global contact centre operations
Humanoid robots are getting faster and that matters more than it sounds
Figure shared a short clip of its latest humanoid robot showing running, turning, and fast acceleration. That's notable because most humanoid robots still move at slow walking speeds.
Some published speeds in the sector include:
- Sanctuary AI Phoenix at about 3 mph
- Unitree H1 in a similar range
- Agility Robotics Digit at about 3 to 4 mph
- earlier estimates for Figure 03 at about 2.7 mph
The new clip suggests a jump toward the upper end of jogging speed, with both feet leaving the ground at the same time. It also shows quick starts, smooth deceleration, and clean directional changes.
This is one of those "small" capability shifts that unlocks a lot. Speed plus stability is what starts to make robots useful outside controlled demos. And the sector is moving quickly, especially with new momentum across China and the ongoing pace at Boston Dynamics.
Australia is building real AI infrastructure and that changes what's possible locally
NEXTDC has announced plans to build an AI campus and GPU supercluster in Western Sydney at its S7 data centre in Eastern Creek. The project is planned to support OpenAI services in Australia and is positioned around local governance needs and alignment to Australia's SOCI framework.
Key points mentioned:
- S7 planned to deliver 550MW capacity when complete
- first phase expected in the second half of 2027
- first phase focused on a sovereign AI campus with high security and resilience standards
- investment expected to create jobs and expand digital infrastructure capability
This is the part people often miss. AI adoption isn't only about models. It's about infrastructure, inference cost, latency, governance, and access. More local capacity changes the economics for everyone.
The AI market is entering a pressure phase and the easy assumptions are fading
I've been using OpenAI, Anthropic, and Gemini for different tasks and it's striking how varied the ecosystem has become.
Recent reporting suggests the sector is entering a period of major pressure as competition intensifies across both hardware and model development.
A few themes stand out:
- Nvidia's dominance in AI hardware is facing real competition as hyperscalers look for alternatives that reduce cost and energy use
- Google's chip strategy and the growth of Gemini are creating pressure for both OpenAI and Nvidia
- OpenAI is accelerating model improvements while delaying new products as it manages rising investment needs
- big platforms are diversifying supply and building their own silicon to spread risk
- the scale of capital required is becoming a deciding factor in who can sustain momentum
This feels like the end of the "one obvious winner" era. The market is getting more fragmented, but also more dynamic. And what happens next will depend on who turns investment into durable capability, not just flashy demos.
Read more: It's a code red moment for the dominant AI players
The point I keep coming back to
AI at work is now normal. But transformation is still rare.
The organisations getting real value aren't just giving staff a chatbot. They're doing the slow, unglamorous work of changing culture, redesigning workflows, building internal capability, and making AI part of the operating system.
Most companies are still stuck in the middle: lots of excitement, some productivity wins, but not a lot of structural change.
That's fine. It's early.
But the gap between "we're using AI" and "we're transformed by AI" is where the real work is going to be done over the next few years.