Hello, I’m Elliot Gray from RealityBreaks.
For the past few years, much of the artificial-intelligence debate has revolved around demonstrations.
Which chatbot gives the smartest answer?
Which model writes the best code?
Which image generator creates the most realistic photograph?
This week brought something rather more important.
We are beginning to see evidence of what happens when AI moves beyond demonstrations and becomes part of the economy, cybersecurity, online shopping and everyday business operations.
Britain’s latest economic figures contain signs that technology and AI-related activity are contributing meaningfully to growth.
Google has released yet another faster and cheaper Gemini model aimed specifically at coding and AI agents.
Taiwan has confirmed an overseas cyberattack in which AI agents were reportedly used together like a coordinated hacking team.
And the battle over who controls AI is changing again as American companies respond to the growing popularity of inexpensive Chinese open-weight models.
For individuals and small businesses, these developments matter far more than another impressive benchmark score.
Let’s look at them in plain English.
1. Google Launches Gemini 3.7 Flash — Just Three Weeks After Its Previous Version
Google released Gemini 3.7 Flash on 13 August, only around three weeks after Gemini 3.6 Flash.
Google describes the new version as a model designed particularly for coding, software engineering and agentic workflows. It says the system has improved at debugging, resolving software problems, creating production-ready code and following complicated instructions. Reuters
Perhaps equally significant is the price.
Google is initially charging $0.75 per million input tokens and $3.75 per million output tokens, around half the original price of Gemini 3.6 Flash. Google is also using the model in Spark, its personal AI agent capable of carrying out ongoing tasks on a user’s behalf. Reuters
Beginner-Friendly Explanation
A token is roughly a small piece of text that an AI processes.
You do not normally need to think about tokens when casually using an AI chatbot, but businesses using thousands or millions of AI requests certainly do.
The important story here is therefore not simply that Google has released another Gemini.
It is that capable AI is becoming cheaper extremely quickly.
At the same time, these models are increasingly being designed for agents—AI systems that can use tools and carry out several steps rather than merely provide an answer.
Why This Matters
The economics of AI are changing.
For businesses processing large volumes of emails, documents, website enquiries or other repetitive work, cutting the cost of each AI task by half can transform whether automation is commercially viable.
The extraordinary speed of Google’s release cycle also shows why businesses should avoid becoming obsessed with having the “latest” model.
Three weeks is barely enough time to properly deploy and evaluate a business system before a successor appears.
Practical Takeaways
For Individuals
Don’t feel pressured to change AI tools every time a new version appears.
Judge them against what you actually do.
For everyday writing, summarising, research and planning, speed and reliability may be more valuable than winning a benchmark competition.
For Small and Medium-Sized Businesses
Start measuring cost per useful task, rather than asking which model is theoretically smartest.
For example:
- How much does it cost to process 1,000 customer enquiries?
- How much staff time is saved?
- How often does the result need correcting?
- Could a cheaper model perform the routine part?
- Which decisions still require a human?
This is likely to become much more important as AI moves into high-volume business processes.
2. Taiwan Confirms an AI-Assisted Cyberattack on Government Systems
One of the week’s most important security stories came from Taiwan.
Taiwan’s Ministry of Digital Affairs confirmed on 13 August that government agencies had been targeted the previous month by an overseas cyberattack involving artificial intelligence. WTVB
Cybersecurity company Dream said the attackers used multiple AI agents working together to perform different hacking tasks. Reports linked to its investigation said credentials were extracted, personnel information was stolen and other sensitive systems were investigated. Taiwan said the affected organisations had responded to the incident. The Straits Times
Taiwan did not publicly attribute responsibility for the attack to a particular country. WTVB
Beginner-Friendly Explanation
We have previously discussed AI agents being used by legitimate businesses.
For example, one agent might research information while another analyses it and a third prepares a report.
Cyber criminals can use exactly the same principle.
Instead of one human hacker manually performing every step, several AI systems can be assigned different tasks, such as:
- looking for vulnerable systems;
- testing passwords;
- analysing stolen information;
- writing attack code;
- identifying new targets.
Humans can still direct the overall operation, but AI allows parts of it to happen considerably faster.
Why This Matters
This demonstrates the other side of AI productivity.
The same technology that allows a business to complete more work with fewer repetitive manual steps can allow criminals to do the same thing.
Cybersecurity therefore becomes more important as AI improves—not less.
For smaller businesses, this is particularly significant because criminals do not only attack governments and multinational corporations.
SMEs are often attractive targets because their security may be weaker.
Practical Takeaways
For Individuals
Basic cybersecurity habits are becoming even more important:
- use different passwords for important accounts;
- use a password manager;
- enable two-factor authentication;
- be suspicious of urgent messages;
- verify unexpected payment requests separately.
AI can make fraudulent emails and messages substantially more convincing than the badly written scams many people learned to recognise in the past.
For Small and Medium-Sized Businesses
Assume phishing attempts will increasingly be well-written, personalised and convincing.
Train staff not to rely on spelling mistakes as the main sign of fraud.
Introduce secondary verification for:
- changed bank details;
- large payments;
- password-reset requests;
- confidential information requests;
- unusual instructions supposedly from directors or suppliers.
3. Britain Is Starting to See AI Show Up in the Economic Numbers
This was perhaps the most interesting story of the week from a UK perspective.
Official figures showed the UK economy grew 0.4% during the second quarter of 2026. According to Reuters’ analysis, the information and communications sector contributed almost half of that expansion. Reuters
Computer programming, consultancy and related activities—areas encompassing much of Britain’s growing AI industry—expanded by 3.7% during the quarter, following growth of 3.8% in the previous quarter. 1470 & 100.3 WMBD
This does not mean AI single-handedly produced Britain’s economic growth. But Reuters notes that the investment boom and productivity improvements associated with AI are beginning to become visible in economic activity. Reuters
Beginner-Friendly Explanation
For years, people have claimed that AI will improve productivity.
Productivity simply means producing more useful work from the same amount of time, money or labour.
Until recently, much of the AI productivity debate relied on predictions or individual company examples.
Now we are beginning to see hints of it in broader economic figures.
Why This Matters
This moves the AI debate beyond hype.
If organisations really are becoming more productive, AI can eventually influence:
- company profitability;
- wages;
- economic growth;
- competitiveness;
- employment patterns;
- government tax revenues.
The effects will not all be positive or equally distributed.
Some jobs will change. Some businesses will benefit faster than others.
But businesses that completely ignore productivity-enhancing technology risk falling behind competitors that learn to use it effectively.
Practical Takeaways
For Individuals
Focus on AI-assisted productivity, rather than simply learning features.
Ask yourself:
Which part of my working day takes an hour that AI could help me complete in 30 minutes?
That might include research, drafting, data analysis, preparation, organisation or repetitive administration.
Keep a record of the time saved. That is evidence of your increasing value—not merely evidence that you have learned to use another piece of software.
For Small and Medium-Sized Businesses
Pick one business process and measure it.
Record how long it takes without AI and how long it takes with AI, including the time required for human checking.
A business that saves 20 minutes on a task performed once a month has achieved very little.
A business that saves 20 minutes on something performed 100 times each week has found something important.
4. Chinese Open-Weight Models Force Silicon Valley to Change Strategy
One of the most significant competitive shifts in AI continued this week.
Chinese companies including Moonshot, Z.ai and DeepSeek have attracted growing international attention with capable open-weight AI models that can be customised and operated more independently than traditional closed systems.
Reuters reported on 12 August that American companies are now responding. Meta is returning strongly to open-weight models, while Nvidia is developing and releasing its own systems as US companies attempt to provide alternatives to increasingly capable Chinese technology. Reuters
DeepSeek added to that competition on 13 August by formally releasing V4 Pro, a higher-performance version of its V4 family. Reuters
Beginner-Friendly Explanation
Most people use a closed AI model.
You visit ChatGPT, Gemini or another service and use the model on the company’s servers.
An open-weight model gives developers access to the trained numerical parameters that make the model work.
That can allow an organisation to run the model elsewhere and customise how it operates.
This is particularly attractive to businesses that want greater control over their data, costs or software environment.
Why This Matters
For several years, the assumption was that the AI industry might eventually be dominated by a handful of enormous American providers.
That now looks far less certain.
Cheaper Chinese models have demonstrated that strong AI performance can be delivered through different business models and at much lower prices.
Competition is good news for customers.
It can produce:
- lower prices;
- more choice;
- faster innovation;
- specialised business models;
- less dependence on one supplier.
There are also legitimate questions about security, training data and trust, particularly when models come from jurisdictions with different legal systems. Reuters notes that concerns about data security remain an obstacle to adoption of some Chinese models by US companies. Reuters
Practical Takeaways
For Individuals
Do not assume the best AI automatically comes from the most familiar brand.
But do not choose solely on price or benchmark results either.
Consider privacy, reliability, terms of service and where your data is processed.
For Small and Medium-Sized Businesses
Avoid hard-wiring an important business process to a single model if you can avoid it.
Keep:
- prompts;
- templates;
- workflow documentation;
- source data;
- final business records
under your own control.
That makes switching provider much easier if prices, quality or availability change.
5. AI Shopping Is Creating a New Problem for Retailers: Who Owns the Customer?
Last week’s RealityBreaks roundup looked at evidence that shoppers are increasingly discovering products through AI.
This week’s development takes that one stage further.
Dutch payment company Adyen warned on 13 August that AI shopping assistants could increasingly handle the complete customer journey—from recommending a product and choosing a retailer to initiating the payment. Reuters
That could be wonderfully convenient for consumers.
For retailers, however, it creates a problem.
If the shopper interacts primarily with an AI assistant rather than the retailer, the retailer may lose the direct customer relationship that generates repeat purchases and loyalty.
Adyen said merchants are therefore becoming increasingly focused on retaining those direct relationships as AI intermediaries become more influential. Reuters
Beginner-Friendly Explanation
Imagine telling an AI:
“Find me a good pair of waterproof walking boots under £100 and buy the best option.”
The AI could eventually:
- search multiple websites;
- compare products;
- read reviews;
- choose a retailer;
- place the order.
You may barely visit the retailer’s website.
That means the AI—not Google and perhaps not even the retailer—controls the discovery process.
Why This Matters
This could fundamentally alter online marketing.
Businesses have spent years building strategies around:
- Google rankings;
- social-media followers;
- email lists;
- paid advertising;
- repeat website visitors.
AI assistants may become another powerful gatekeeper.
Being recommended by AI therefore becomes important.
But so does ensuring the customer has a reason to come directly back to your business afterwards.
Practical Takeaways
For Individuals
AI shopping assistants could be extremely useful for comparing complicated products.
But keep final control.
Check price, delivery, returns, warranty and seller reputation before completing an important purchase.
For Small and Medium-Sized Businesses
There are two strategies worth pursuing simultaneously.
First, make your website easy for AI systems to understand. Clearly explain products, prices, specifications, availability, returns and frequently asked questions.
Second, strengthen the reason customers return directly to you.
That might include:
- useful email content;
- excellent after-sales support;
- loyalty benefits;
- warranties;
- personal expertise;
- members-only information;
- genuine customer relationships.
In an AI-mediated shopping world, brand loyalty may become more valuable, not less.
6. Most Japanese Companies Still Haven’t Fully Adopted AI
Not every country or company is racing ahead.
A Reuters survey published on 12 August found that more than 80% of Japanese companies have either adopted AI only in limited parts of their operations or have not adopted it at all. Reuters
Around 60% said they were using AI only in certain areas, while just 16% described AI as an integral company-wide tool. Eighteen percent had not yet decided whether to adopt it and 6% were not considering adoption. Reuters
That is particularly striking because Japan is one of the world’s most technologically advanced economies.
Beginner-Friendly Explanation
There is an enormous difference between having access to AI and actually changing how an organisation works because of AI.
A business may say it uses AI because a few employees occasionally ask a chatbot to rewrite an email.
That is very different from systematically using AI across:
- customer support;
- sales;
- research;
- administration;
- operations;
- finance;
- internal knowledge.
Most businesses worldwide are probably still somewhere between those two stages.
Why This Matters
This story should reassure small-business owners who feel they have already been left behind.
They haven’t.
Even major corporations in highly developed economies are still figuring this out.
The opportunity remains considerable.
It also demonstrates why buying an AI subscription is not an AI strategy.
People need training. Workflows need redesigning. Rules need establishing. Results need measuring.
Practical Takeaways
For Individuals
Do not measure your progress against AI experts on social media.
If you can use AI confidently for several genuinely useful tasks, understand its limitations and know how to check its answers, you are already developing valuable skills.
For Small and Medium-Sized Businesses
You do not need a five-year AI transformation programme.
Begin with three questions:
- Where do we waste the most staff time?
- What repetitive work could AI assist with?
- What information do employees repeatedly need to find?
Solve one worthwhile problem.
Then solve another.
That is far more effective than introducing AI everywhere simply because competitors are talking about it.
Elliot Gray’s Closing Perspective
This week gives us one of the clearest pictures yet of where artificial intelligence is heading.
Google is making capable AI cheaper.
China is forcing American companies to rethink how open their models should be.
Britain is beginning to see signs of AI-related activity in actual economic growth.
Consumers are beginning to let AI influence what they buy.
Cyber attackers are using agents to automate parts of hacking operations.
And at the same time, more than 80% of companies in one of the world’s most advanced economies still haven’t fully integrated AI.
That last point is worth remembering.
We hear so much about the speed of AI that it is easy to assume everyone else has already mastered it.
They haven’t.
For individuals, there is still time to learn.
For small and medium-sized businesses, there is still time to gain an advantage.
But the useful question is no longer:
“Should we use AI?”
A better question is:
“Which parts of what we already do could AI genuinely make better?”
Start there.
Choose a real problem.
Test the technology.
Measure the result.
Protect your information.
Keep a responsible human involved.
And expand what genuinely works.
That approach doesn’t require a Silicon Valley budget or a computer-science degree.
It requires curiosity, sensible judgement and a willingness to learn.
And those remain very human advantages.
Call to Action
Want practical artificial-intelligence guidance without the jargon, panic or exaggerated promises?
Visit the RealityBreaks AI Blog for weekly AI news roundups, beginner-friendly tutorials, creative AI guides and practical advice for using artificial intelligence safely at home and in business.
And if you’re over 50 and concerned that technology is moving faster than you can keep up, take a look at AI After 50: The Plain-English Guide to Using Artificial Intelligence. It is designed to help you build practical AI confidence step by step, without coding or technical jargon.