The three AI names I have heard most this week are Sol, Fable and Codex.
Which would be helpful if anybody explained whether they are three models, three products or the cast of a new children’s television programme.
They are not the same thing. They are not priced the same way. And a small business owner does not need all three.
Meanwhile, Canva made it possible to build an interactive website or calculator without coding. Google put Canva inside Search. Congress held an entire hearing about AI for small business. Then the AI companies gave their agents the keys, the company card and a browser.
Which feels like the natural next step, in the same way giving a toddler an espresso and access to Amazon feels like the natural next step.
The biggest AI news this week is not another chatbot getting 4.7 percent better at a test nobody outside a laboratory has heard of. It is that the best models are becoming easier for small businesses to use for finished work, while agents are being connected to passwords, payments, websites, private research and real accounts.
That changes the question for small business owners.
It is no longer, “What can AI write for me?”
It is, “What can I safely let it do?”
I went through the announcements, research papers and product launches from 13 to 19 July 2026. I kept the ones that change something for a real business and left the rest in the press release drawer, where it can talk about paradigm shifts to itself.
For every story, I have included what happened, what it means and what I would do with it.
The short version
The AI news this week for small business is about getting more useful work from fewer tools. GPT-5.6 Sol is OpenAI’s strongest general model, Claude Fable 5 is formidable on long and complex work, and Codex is the working environment that can use models to build and change things. Canva Code 2.0 can now create editable landing pages, calculators and mini apps for free. Google is connecting Search to Canva and other services. The less glamorous stories matter too: AI agents can use passwords without seeing them, websites built for AI agents completed shopping tasks almost twice as reliably, and OpenAI published a useful way to measure whether any of this is making or saving you money.
If you only read three sections, read the Sol vs Fable vs Codex guide, Canva Code 2.0 and the AI ROI scorecard. Those are the ones most likely to change what a small business owner does on Monday.
The AI news this week at a glance
Click any story to jump straight to it:
- GPT-5.6 Sol vs Claude Fable 5 vs Codex: what should a small business use?
- Canva Code 2.0 can build the lead magnet instead of another PDF
- Congress put AI for small business on the main stage
- Google turned NotebookLM into Gemini Notebook
- OpenAI finally gave businesses a sensible AI ROI scorecard
- Claude can now use your passwords without seeing them
- Visa, Mastercard, Google and others backed payments for AI agents
- A new study showed why ordinary websites fail AI buying agents
- Cloudflare says bots now generate 57 percent of web requests
- Claude for Teachers showed where useful AI products are heading
- Perplexity showed how long-running agents can work safely
- NVIDIA pushed robots and physical AI closer to the mainstream
- The pattern beneath all twelve stories
1. GPT-5.6 Sol vs Claude Fable 5 vs Codex: what should a small business use?
Let us clear up the names first.
GPT-5.6 is OpenAI’s new model family. Sol is the most capable version, Terra is the balanced everyday version and Luna is the fastest and cheapest.
Claude Fable 5 is Anthropic’s most capable generally available Claude model. It is especially strong on long, complicated knowledge work and software tasks. Its included access has been extended and changed several times, which has produced the sort of pricing clarity normally associated with trying to split a restaurant bill between fourteen people.
Codex is different. Codex is OpenAI’s agent and working environment. It uses models, including GPT-5.6, to inspect files, research, write documents, analyse data, build websites and carry out longer jobs. Comparing Codex with Sol is like comparing Microsoft Word with the person writing inside it.
OpenAI also launched the Codex Micro this week, a $230 desktop keypad for controlling coding agents. It has keys for common Codex actions, a dial for changing the reasoning level and a light showing what an agent is doing.
You almost certainly do not need one.
The important part is not the keypad. It is the assumption underneath it: people will soon manage several AI workers at once and will need an easy way to see what they are doing, stop them and approve important actions.
Which one should a small business use?
| Your job | Best starting point | Why |
|---|---|---|
| Everyday emails, summaries and admin | GPT-5.6 Terra on medium effort | Strong enough for routine work without paying for maximum thinking |
| Important strategy, financial analysis or a difficult decision | GPT-5.6 Sol on high effort | Better reasoning is worth paying for when the decision matters |
| Long reports, nuanced writing and work that must hold lots of context | Claude Fable 5 | This is where Fable’s long-form strength earns its keep |
| Building or fixing a website, tool, spreadsheet workflow or automation | Codex using Terra or Sol | Codex can work across files and produce the finished asset, not merely tell you how |
| High-volume, low-risk tasks | GPT-5.6 Luna | Speed and cost matter more than squeezing out the last five percent of quality |
That does not mean the table is a law handed down on a stone tablet. Run the same real task through two tools and judge the finished result, correction time and price.
Benchmarks are useful. Your invoice, your deadline and the amount of rewriting required are more useful.
Use it in your business this week
Take one job you do every week and test it three ways:
- Your current model and setting
- A stronger model with higher effort
- A cheaper model with lower effort
Record the time, quality, correction needed and total cost. Keep the cheapest option that clears your quality bar.
For most small businesses, the answer will not be “buy all of them”. It will be one everyday tool, plus temporary access to a stronger model when the work justifies it.
My guide to the best AI tools for small business owners gives you the broader stack. If you are wondering whether ChatGPT or Claude is enough without outside help, read my AI consultant vs ChatGPT cost comparison.
2. Canva Code 2.0 can build the lead magnet instead of another PDF
Canva Code 2.0 became available to everyone this week, including free users.
It lets you describe a website, mini app, quiz, calculator, game or interactive landing page in plain English. Canva builds it, then lets you edit the colours, type, images and text inside the same drag-and-drop editor people already know.
You can also import HTML made elsewhere and turn it into an editable Canva project. Canva says code generation is 75 percent faster and the median time from prompt to published output is 30 percent faster. More than six million Canva Code sites have already been created.
This is more relevant to a small business than another model adding three points to a reasoning benchmark.
What this means for your business
Most lead magnets are still PDFs.
Some are excellent. Some are twelve pages of information the reader could have found by asking Google while waiting for the kettle.
An interactive tool gives the reader a result that belongs to them. A calculator can estimate their wasted ad spend. A quiz can diagnose which service they need. A planner can turn five answers into a ninety-day action list. A checker can show whether AI systems can find their business.
That is more useful, more shareable and more likely to attract a link than another “ultimate checklist” wearing a stock-photo cover.
It is also a conversion tool. The user invests information, receives a personalised answer and has a natural reason to take the next step.
Five things a small business could build
- A price or savings calculator
- A product finder that recommends the right option
- A readiness score for the problem you solve
- A client onboarding planner
- A short diagnostic that ends with a tailored action plan
Start with a narrow question. “How much could this save me?” works. “Please recreate my entire business inside Canva by tea time” is a less promising first brief.
Use it in your business this week
Take your most downloaded PDF and ask what result the reader wanted from it.
Then prompt Canva Code:
“Build a simple mobile-friendly calculator for [audience]. Ask five questions about [problem]. Show a personalised score, three recommendations and one clear next action. Use [brand colours]. Do not submit or store personal information.”
Check the maths, mobile layout, privacy and call to action before publishing. Canva has made building easier. It has not made checking optional.
Google also connected Canva directly to AI Mode in Search this week. US users can ask for flyer or design ideas while searching and open the result in Canva. Instacart and YouTube Music are connected too.
The larger signal is that AI search is becoming an action layer. People will research, design and shop without visiting the old sequence of ten separate websites.
That makes your own interactive tools and AI visibility more valuable. My free AI Visibility Checker is the kind of tool I mean. I have also explained how to run a fifteen-minute AI visibility audit if you want to see what ChatGPT, Gemini, Claude and Perplexity say about your business.
3. Congress put AI for small business on the main stage
On 14 July, the US House Committee on Small Business held a hearing called AI on Main Street.
That sounds dry enough to remove moisture from a bathroom wall, but the evidence was useful.
Jordan Crenshaw from the US Chamber of Commerce told the committee that 58 percent of small businesses in its national survey now use generative AI. Adoption rises to around 70 percent among smaller technology firms.
Anthony Qaiyum, president of a Chicago retailer founded in 1875, said AI had helped his business improve its capacity and hire ten new employees. The committee’s conclusion was not that small businesses cannot access AI. It was that many do not have the training or knowledge to use it well.
That distinction matters.
Small businesses are no longer behind because they have not heard of ChatGPT. They are behind when they use it for the occasional social caption while a competitor connects it to customer service, inventory, sales follow-up and decision-making.
What this means for your business
Adoption is no longer an advantage by itself.
Using AI well is the advantage.
The small businesses pulling ahead are doing three things:
- Applying AI to a repeatable job rather than random prompts
- Giving it real business context rather than a blank chat
- Measuring the finished result rather than celebrating how much content appeared
This is why the model comparison matters less than the workflow around it. Fable, Sol and Codex can all produce expensive nonsense when the task is vague. A cheaper model with the right documents, boundaries and quality check can produce something worth keeping.
Use it in your business this week
Write down the five jobs that consume the most time in your business.
Circle the one that is repeated, rules-based and easy to check. That is your next AI workflow.
Do not begin with your most sensitive data or the task that could refund every customer by accident. Begin with something dull, frequent and reversible.
My practical AI implementation roadmap shows how to move from testing to a working system. If generic answers are the problem, build a private AI assistant using your own business documents.
4. Google turned NotebookLM into Gemini Notebook
Google renamed NotebookLM to Gemini Notebook this week.
Normally, a rename is not news. It is a meeting that got out of hand.
This one came with something useful.
Google has started giving every notebook a secure cloud computer. Gemini Notebook can write and run code against the material inside your notebook, which means it can move beyond summarising documents and start analysing the information in them.
Google says more than 30 million people and 600,000 organisations already use the product. The new code-running feature is available to Google AI Ultra users and some Workspace business customers now, with Pro access rolling out on the web over the next few weeks. Notebooks already sync with the Gemini app and will eventually appear inside AI Mode in Google Search.
What this means for your business
NotebookLM was already good at answering questions about a pile of source documents without wandering off and inventing a cousin who works at Deloitte.
Giving it a computer makes it much more interesting.
Imagine adding twelve months of sales reports, campaign results, customer interviews and support tickets to one notebook. Instead of asking for a summary, you could ask it to calculate which customer type stays longest, chart the complaints that grew fastest, or compare the promises in your marketing with what customers praise after buying.
The sources, the analysis and the output stay together.
That is a much better business tool than a blank chat box because it starts with your evidence, not the internet’s average answer.
Use it in your business this week
Create one notebook for one decision you are already trying to make.
Do not upload your entire digital attic. Give it a clean box of evidence:
- Your last six monthly reports
- Your five best and five worst sales calls
- Your customer survey responses
- Your current offer and three competitor offers
Then ask three questions:
- What pattern would I miss if I read these one by one?
- Which conclusion is supported by the strongest evidence?
- What should I stop, start or test next?
If you want a wider system for putting AI to work without creating a mess, read my practical AI implementation roadmap.
5. OpenAI finally gave businesses a sensible AI ROI scorecard
AI companies love talking about tokens.
Small business owners do not sell tokens.
We sell work, decisions, products, appointments, results and, on difficult Tuesdays, the remaining scraps of our patience.
OpenAI published A scorecard for the AI age on 17 July. Its central idea is “useful intelligence per dollar”. In other words, stop measuring how much AI your team uses and start measuring how much worthwhile work gets finished.
The scorecard asks four questions:
- Is AI completing work that matters?
- What does each successful task cost?
- Can people depend on the result?
- Does each AI dollar produce more value as usage grows?
Two days earlier, OpenAI published a related five-step framework for managing AI investment. It recommends tracking usage and spend, measuring cost per accepted result, putting controls around agent workflows before scaling them, funding repeatable workflows and matching capacity to proven demand.
OpenAI also reported that the price per million tokens fell 97 percent from GPT-4 to GPT-5.4. Its own benchmark found GPT-5.6 used 54 percent fewer output tokens and 57 percent less time per task. Useful, but the more important point is buried underneath: the cheapest model is not always the cheapest result. A cheap model that needs four attempts and half an hour of correction can cost more than an expensive model that gets the job right once.
The five-number AI scorecard I would use
Forget the twenty-seven-tab dashboard. Track these five numbers for each AI workflow:
| Measure | What to record | Example |
|---|---|---|
| Completed outcomes | Work accepted without being redone | 42 customer replies approved |
| Human minutes saved | Old time minus new review time | 310 minutes saved |
| Total cost | Software, model usage and human review | $38.40 |
| Dependability | Accepted outcomes divided by all attempts | 42 out of 50, or 84% |
| Business value | Revenue created, cost avoided or capacity freed | Two extra sales calls booked |
Then calculate two figures:
Cost per accepted outcome = total cost divided by completed outcomes
Return on AI = business value minus total cost, divided by total cost
You do not need perfect accounting. You need enough truth to spot the workflow that saves six hours and the shiny subscription that has produced three limericks and nothing else.
Use it in your business this week
Pick the AI workflow you use most.
Email drafting. Content repurposing. Meeting notes. Lead research. Customer support.
Measure it for seven days using the five numbers above. At the end of the week, choose one of four actions:
- Keep it because it creates measurable value.
- Improve it because the idea works but the error rate is too high.
- Move it to a cheaper model because the quality bar is easy to meet.
- Cancel it because you have been paying for the feeling of progress.
I have written more about this in the real cost of AI tools for small businesses and how to measure ROI on AI.
6. Claude can now use your passwords without seeing them
This is the story I would have called impossible a year ago.
1Password for Claude launched on 16 July. It lets Claude log into websites using credentials stored in 1Password, but the password and one-time code never enter Claude’s context, memory or Anthropic’s systems.
When Claude reaches a login, 1Password shows you which credential it wants and why. You approve the request with biometrics. 1Password injects the credential into the page through a separate secure channel. Claude knows it logged in, but it never sees what it typed.
Access lasts for that task and does not carry into another session. When a compatible agent controls the browser, 1Password’s new Agentic Mode locks the vault so the agent can reach only the items you approved. If a form submission fails, 1Password clears the filled values before returning control.
It is available now for 1Password users on Mac and works with the Claude desktop and browser extensions.
What this means for your business
The bottleneck for browser agents has never only been intelligence.
It has been trust.
You can ask an agent to reconcile Stripe, check an advertising account or update a customer record, but sooner or later it reaches a login screen. At that point, you either expose the password or stop the automation and do it yourself.
This integration offers a third option: permission to use a secret without permission to know the secret.
That distinction will matter everywhere. Today it is Claude and 1Password. Soon it will be every serious business agent and every system holding money, customer data or publishing rights.
Use it in your business this week
Do not celebrate by giving an agent the keys to every account you own.
Start with one read-only task in one low-risk account. Ask it to retrieve information, not change anything. Approve one credential for one session. Check the result and the activity log.
The safe progression is:
- Read-only access
- Draft or prepare an action
- Human approval
- Reversible write
- Live verification
If a workflow jumps from step one to “let it buy things while I sleep”, the workflow has been designed by someone who enjoys support tickets.
My guide to the AI policy a small business needs gives you the rules to set before agents start wandering through company accounts.
7. Visa, Mastercard, Google and others backed payments for AI agents
Last week I wrote about the first attempts to let AI agents pay websites.
This week, it stopped looking like a side project.
On 14 July, the Linux Foundation announced the operational launch of the x402 Foundation. The x402 protocol adds payments to ordinary web requests, so an AI agent, API or application can pay for access as easily as it requests data.
Forty organisations have joined. Premier members include American Express, AWS, Cloudflare, Coinbase, Google, Mastercard, Shopify, Stripe and Visa.
That list is why this matters.
When the card networks, payment processors, cloud companies and ecommerce platforms sit at the same table, they are not discussing whether agents will spend money. They are discussing how.
What this means for your business
Today’s website assumes the buyer is a human with eyes, a mouse and a willingness to hunt through six tabs for the price.
Tomorrow’s buyer may send an agent with instructions like:
“Find a consultant with at least fifteen years of experience, under this budget, available in August, with evidence they have done this before.”
The agent will compare options, reject the vague ones and may complete part of the transaction.
If your offer, availability, proof and terms are hard for a machine to understand, you may never reach the human’s shortlist.
Use it in your business this week
You do not need to accept agent payments this week.
You do need to make your commercial information easier to read:
- Give every service a plain-English description.
- Put concrete inclusions and exclusions on the page.
- Use consistent names for products and packages.
- Make the next action obvious.
- Add structured data where it fits.
- Keep dates, availability and terms current.
This is not about decorating your website for robots. It is about removing the ambiguity that frustrates humans too.
8. A new study showed why ordinary websites fail AI buying agents
This may be the most useful research paper I read all week.
Researchers Said Elnaffar and Farzad Rashidi published a preprint called Designing Agent-Ready Websites for AI Web Agents on 13 July.
They built two versions of the same ecommerce website. The catalogue, prices, stock and shopping tasks were identical. One version was designed mainly for humans. The other added clearer structure, action cues, evidence and signals showing whether information was current.
They tested five tasks with three browser-agent models over 300 runs.
The ordinary website produced 74 successful runs out of 150, a strict success rate of 49.3 percent.
The agent-ready version produced 134 successful runs out of 150, or 89.3 percent.
Partial outcomes fell from 43 to 3. Average steps dropped from 9.31 to 6.49.
It is a controlled prototype and a preprint, not the last word on the subject. But the difference is too large to ignore.
What made the agent-ready website work
The researchers grouped the improvements into three areas:
Interpretability: Can the agent understand the products, facts, relationships and page structure?
Executability: Can it tell what action to take and complete it without guessing?
Decision reliability: Can it verify the evidence, constraints and freshness of the information before recommending something?
That gives us a practical website audit.
The 12-point agent-ready website checklist
Check one important offer page against these questions:
- Does the first paragraph say exactly what this is and who it is for?
- Are product and service names used consistently everywhere?
- Are prices, limits and inclusions written as text rather than hidden in graphics?
- Can a visitor compare options in a table?
- Does every button describe the action it performs?
- Are forms labelled clearly?
- Are availability, stock or delivery times explicit?
- Does the page show when time-sensitive information was updated?
- Are claims supported by named evidence?
- Is contact and business identity information easy to find?
- Does structured data accurately match the visible page?
- Can someone complete the next step without opening another mystery tab?
If you score below nine, an AI agent is probably not your only confused visitor.
Use it in your business this week
Run your most valuable service or product page through an AI browser agent and give it a real buying task.
Do not ask, “Is this page good?” That produces compliments, which are free and worth roughly what you paid.
Ask:
“Find the price, explain exactly what is included, identify who this is for, tell me what proof supports the offer, and complete the next step without submitting anything. Record every point where you had to guess.”
Every guess is a conversion leak.
This is the next layer after being mentioned in AI answers. First the AI must find you. Then it must understand you. Then it must be able to act. You can check the first part with my free AI Visibility Checker and use my guide to getting your business cited by ChatGPT for the content side.
9. Cloudflare says bots now generate 57 percent of web requests
Cloudflare launched a new bot-detection product called Precursor this week.
The product announcement is less interesting than the number inside it.
Cloudflare says automated bot traffic now generates roughly 57 percent of all web requests, overtaking human activity for the first time.
Its new system watches behaviour across a complete browsing session instead of presenting one CAPTCHA at the door. A modern bot can fake a click. It has a harder time faking an entire human journey.
What this means for your business
“Bot” no longer means one thing.
Some bots scrape your content. Some check prices. Some monitor stock. Some search on behalf of a potential customer. Some attack forms, steal accounts or run up server costs. Blocking all automation is becoming as blunt as locking a shop because one customer looked suspicious.
Businesses need to distinguish useful agents from abusive automation.
That affects security, analytics and marketing. A traffic spike may not be a sudden burst of human interest. A zero-conversion visit may still be an AI agent collecting evidence before recommending you elsewhere. Your old dashboard does not tell that story neatly.
Use it in your business this week
Look at three things:
- Your server or Cloudflare bot reports
- Unusual spikes in form attempts, page requests or login failures
- Whether legitimate AI search crawlers can still access the pages you want cited
Do not solve a security problem by disappearing from AI search. Separate malicious automation, training crawlers and search or user-triggered agents before you change the rules.
My guide to getting recommended by ChatGPT and Perplexity explains why crawler access and citation visibility need to be treated as a business decision, not a panic button.
10. Claude for Teachers showed where useful AI products are heading
Anthropic launched Claude for Teachers on 14 July, giving verified US K-12 educators free premium access for a year if they sign up by 30 June 2027.
The product connects Claude to educational standards across all fifty states, evidence-based curricula and classroom tools. It includes teaching skills for lesson planning, adapting material for different ability levels and analysing class data. Anthropic says teacher data is not used for model training and has added terms designed for K-12 privacy and FERPA compliance.
What this means for your business
The interesting bit is not that teachers get free Claude.
It is how Anthropic packaged it.
This is not a generic chatbot with “act as a teacher” pasted into the prompt. It has trusted source material, sector-specific tools, tested workflows, privacy rules and a clear user group.
That is what useful vertical AI looks like.
The same pattern applies to accountants, estate agents, consultants, clinics, charities and agencies. The winning product is rarely “AI for everyone”. It is AI that understands the documents, rules, language and recurring work of one type of business.
Use it in your business this week
If you are building an AI service, stop describing it by the model underneath.
Customers do not want “a GPT-powered solution”. They want the late invoices chased, the enquiries sorted, the sales calls summarised or the lesson plans finished before midnight.
Write down:
- The person you are building for
- The recurring job eating their time
- The trusted information the AI needs
- The actions it may take
- The actions that need human approval
- The evidence that proves it worked
That is an offer. “AI automation” is a category label wearing a lanyard.
For a practical starting point, see my AI implementation roadmap for small businesses.
11. Perplexity showed how long-running agents can work safely
Perplexity published the technical design behind SPACE, the sandbox system powering millions of Perplexity Computer sessions.
A sandbox gives an agent somewhere isolated to work. It can run code, manipulate files and carry out a long task without being handed the keys to the whole machine.
This matters because the new generation of agents does not answer once and stop. It may research for an hour, create files, test an output, recover from an error and continue. More ability means more ways to make a mess if the boundaries are poor.
What this means for your business
When you delegate work to a person, you do not automatically give them access to the payroll, your private photographs and the button that deletes the website.
Agents need the same common sense.
The safe unit of delegation is a defined workspace with the minimum files, tools, time and permissions required for the job.
Use it in your business this week
Before giving an agent a long-running task, define five boundaries:
- What it can read
- What it can change
- What it can send or publish
- When it must stop for approval
- How you will verify the final state
If those five answers are not written down, you have not delegated a task. You have released a very confident intern into the building after dark.
12. NVIDIA pushed robots and physical AI closer to the mainstream
NVIDIA introduced two smaller Jetson Thor modules on 15 July: the T3000 and T2000.
The T3000 delivers similar inference performance to NVIDIA’s larger T5000 for multimodal workloads, in roughly half the size and power. The T2000 is the lower entry point for visual agents, mobile robots and industrial machines. NVIDIA also released agent skills that optimise memory and deployment, work that previously needed specialist engineers.
The modules are scheduled for the first quarter of 2027, so nobody is putting one in the office kettle this afternoon.
What this means for your business
AI is moving out of the browser.
Running models on a local device matters when speed, privacy or an unreliable internet connection makes the cloud a poor fit. Think cameras that spot defects, machines that respond in real time, retail systems that understand what is happening on the floor, or robots that work alongside people.
For most small businesses, this is “watch, do not buy” news.
The useful signal is that the cost and specialist knowledge needed for physical AI are coming down. The same thing happened with websites, video production and software. First it needs a team. Then it needs one specialist. Eventually it becomes a Tuesday afternoon job.
Use it in your business this week
Look for a physical bottleneck, not a robot-shaped solution.
Where do staff repeatedly inspect, count, move, photograph or check something in the real world? Write that down. Do nothing else yet.
When the tools become affordable, the businesses with a clearly defined problem will move faster than the ones shopping for a robot because it looked good in a keynote.
13. The pattern beneath all twelve stories
Last week’s AI news was about intelligence getting cheaper.
This week’s is about access.
Access to your documents. Access to a computer. Access to websites. Access to credentials. Access to payment rails. Access to physical machines.
That is when AI starts creating serious value.
It is also when bad instructions stop being mildly annoying and start becoming expensive.
A chatbot that misunderstands you writes a duff paragraph. An agent that misunderstands you can change a record, spend money, expose a secret or confidently click the wrong blue button.
The businesses that benefit will not be the ones that grant the most access. They will be the ones that grant the smallest useful amount of access, measure the result and widen the boundary only when the evidence earns it.
What I would do this week
You do not need to try all twelve developments. That would turn your Monday into a low-budget technology conference.
Do these five things instead.
1. Pick one model by the job
Use Terra for routine work, Sol for a difficult high-value decision, Fable for long and nuanced work, and Codex when you need a finished asset built across files. Do not pay the most for every task.
2. Build one tiny interactive tool
Use Canva Code 2.0 to turn an existing checklist or worksheet into a calculator, quiz or diagnostic. Keep the first version narrow enough to test in an afternoon.
3. Measure one AI workflow
Use the five-number scorecard: completed outcomes, human minutes saved, total cost, dependability and business value.
4. Test one important web page as an agent
Ask an AI browser to find the offer, price, proof and next action. Record every guess.
5. Check whether AI can find your business
Run your name through my free AI Visibility Checker. If the engines do not mention you, the clever payment rails and agent-ready buttons will be waiting at a party nobody invited you to.
Five jobs. You do not have to do them all today. Even one is more useful than reading another fifty AI headlines and forgetting forty-nine of them by lunch.
Free resource: The Value-Based Pricing Worksheet Template.
The final word
For three years, AI has been treated like a clever person trapped in a text box.
Ask it something. Copy the answer. Fix the weird bit. Repeat.
That period is ending.
The AI is coming out of the box and into the accounts, websites and workflows where business happens. It can analyse the documents, use the login, make the payment and operate the machine.
Brilliant.
Also, maybe label the cupboards.
The opportunity is enormous, but it does not reward blind trust. It rewards good delegation: the right task, the right evidence, the minimum access, a clear stopping point and a result you can verify.
That is the part of AI implementation nobody puts in the shiny demo. It is also the part that decides whether the shiny demo becomes a useful employee or a small electrical fire.
I will keep reading the week’s AI news and pulling out the developments that change something for your business. You do not need every launch. You need the few that alter what you do next.
This week, that means choosing the right model for the job, building something more useful than another static PDF, measuring value, making your website easier for agents to understand and tightening permissions before you hand over the keys.
Preferably before the toddler orders a forklift.
Related reading: 4 Things to Know Before Hiring an SEO Company for Your Business.
Related reading: 4 Ways To Get More Interest From Your Audience.
Related reading: ai interactive lead magnets small business.
Related reading: Taking Your Business Local Through Social Media.
Related reading: How to Measure Social Media Success Without a Paid Analytics Tool.
Frequently asked questions
What was the most important AI news this week for small businesses?
The most immediately useful development was Canva Code 2.0 becoming available to all users, because a small business can now build an interactive lead magnet, calculator or simple app without hiring a developer. The wider shift is AI moving from producing answers to completing real work across files, websites and connected services.
Is GPT-5.6 Sol better than Claude Fable 5 for small business?
Neither is best for every job. GPT-5.6 Sol is the stronger starting point for difficult reasoning, polished documents, spreadsheets and work that uses tools. Claude Fable 5 is excellent for long, complex analysis and nuanced writing. Test both on a real recurring task and compare the accepted result, correction time and cost.
What is the difference between GPT-5.6 Sol and Codex?
GPT-5.6 Sol is a model. Codex is an agent and working environment that can use models such as Sol, Terra or Luna to research, inspect files, write code and create finished assets. Sol is the brain option. Codex is the place where that brain can carry out the job.
Is Canva Code 2.0 useful for small businesses?
Yes, particularly for simple calculators, quizzes, product finders, interactive lead magnets, landing pages and internal tools. It is available to free users and the output can be edited inside Canva. Check calculations, forms, mobile behaviour, privacy and calls to action before publishing.
How should a small business measure AI ROI?
Track five measures for each repeatable workflow: accepted outcomes, human time saved, total cost, dependability and business value. Calculate cost per accepted outcome rather than cost per token. A more expensive model can be cheaper overall if it completes the task correctly with less correction.
What is an agent-ready website?
An agent-ready website is structured so an AI browser agent can understand the offer, verify information and complete the intended action without guessing. Clear product names, visible prices, comparison tables, descriptive buttons, current information, supporting evidence and accurate structured data all help.
Can Claude now see my passwords?
Not through the new 1Password for Claude integration. After the user approves a credential request, 1Password injects the password or one-time code directly into the page through a separate secure channel. The secret does not enter Claude’s context or memory. Access is limited to the approved task.
What is x402?
x402 is an open payment standard built into ordinary web interactions. It allows AI agents, applications and APIs to send or receive payments programmatically. The Linux Foundation’s x402 Foundation launched operationally in July 2026 with forty member organisations from payments, cloud computing and ecommerce.
Should I redesign my website for AI agents now?
Do not rebuild the entire site for an unproven future. Start by making important commercial pages clearer for humans and machines. Test one real task with an AI browser agent, record where it guesses or fails, and fix those specific points. Clearer information and actions should help human conversion too.
Do I need to use all the new AI tools?
No. Choose tools by workflow, not novelty. Test one repeatable job, define an acceptable result and measure the full cost of achieving it. Keep the tools that create measurable value and cancel the ones that mainly create more tabs.
Get the digest in your inbox
Every Sunday I send one email to business owners, marketers and entrepreneurs who would like the benefit of keeping up with AI without sacrificing their entire personality to it.
You get the week’s useful AI news, the business meaning underneath it and the experiments I am running on my own business.
If you want help deciding where AI belongs in your business, what to automate and what should remain firmly attached to a human, see how I work with businesses.
You can also raid my free AI resources library or check your business with the free AI Visibility Checker.
Related reading: Best Free Digital Business Card Apps: What I’d Download in 2026 and How to Find Trending Hashtags for Free in 2026.
For the bigger picture, see my full guide to AI marketing.
