Claude Sonnet 5 Just Made AI Agents Affordable for Small Business
The week AI agents stopped being a luxury
Hello,
This week’s edition has one story sitting above everything else: Anthropic launched Claude Sonnet 5 on 1 July, and it changes the maths on what capable AI costs a small business. It is reported as Anthropic’s most agentic model yet, meaning it can work through multi-step tasks on its own, at close to flagship performance for a much lower price. If you read last week’s edition, you know I have been watching the price of capability fall for months. This week it fell hard.
We will get to the tools, a quick win you can run this week, and the scams catching Australian business owners right now. First, the headline.
Claude Sonnet 5: agent-level skills at mid-tier prices
Here is what actually happened. Anthropic released Claude Sonnet 5, a model that can use tools like a web browser on its own and follow a process from start to finish without you feeding it each step. Early reporting puts its performance close to Anthropic’s flagship models at a significantly lower cost.
The same week, OpenAI announced GPT-5.6 as three separate models: a flagship, a balanced tier, and a fast, affordable tier starting around US$1 per million tokens. That is a pricing structure, not just a product launch. Both companies are saying the same thing: pick the capability you need, pay for that, nothing more.
62% of small business owners already use AI to run their business, according to US Chamber of Commerce reporting. The question is no longer whether to use it. It is whether you are using it deliberately.
Why does this matter to you? Because the expensive tier was never the barrier. The barrier was that the affordable tier could not finish a job. That gap is closing fast. An assistant that can research a supplier, draft the comparison, and format the summary is now priced like a software subscription, not a consultant.
But capability without structure is just noise. Workflow before automation. Always. If you are not sure where AI fits in your business, start with the guide to implementing AI in your small business before you buy anything.
Tools worth trying this week
Claude Sonnet 5
The new default for capable AI without flagship prices. Drafting, research, and multi-step admin in plain English. If you tried AI a year ago and found it shallow, try again. It is not the same product it was.
Notion AI
Turns messy documents into clean procedures and meeting notes into action items. If your processes live in your head and nowhere else, this is the cheapest fix I know of.
Zapier AI
Describe an automation in plain English and it builds the steps between the apps you already use. No code, no developer, no six-week project.
Canva Magic Studio
Branded designs from a prompt, plus auto-subtitles for video. Keeps your marketing consistent without hiring a designer for every single post.
ElevenLabs
Natural-sounding voiceovers from text. Training videos, phone greetings, walkthroughs. Useful if you would rather not record yourself twelve times to get one clean take.
Industry surveys report owners using AI most for research and brainstorming (51%), content creation (44%) and visuals (37%). Note what is missing: operations. That is where the real hours hide.
Quick win: build a reply library this week
One task, four steps, done by Friday.
First, list the five questions customers ask you most. You already know them. Second, paste real examples into your AI tool, names removed, and ask for a reply template in your tone for each one. Third, edit until they sound like you, then save them somewhere your whole team can reach. Fourth, next Monday, count the time saved. Then add five more.
This works because it is bounded. The AI drafts, you approve, the customer never knows the difference. One client of LDC, a commercial laundry and linen service, had a system built on exactly this principle. During the period the system ran, the client managed over $20,000 in outstanding invoices every week with zero manual follow-up, and over a 12-month period the system returned more than 200 hours of admin to the business. It started with one small, repeatable win. Structure first, then scale.
Watch out for: the scams got personal
Two numbers from this week’s research worth sitting with.
AI-written phishing emails are reported to achieve more than four times the click-through rate of human-written ones. Perfect grammar, your supplier’s tone, your name.
Roughly one in four small business owners has reportedly already been targeted by an AI-driven scam. Most did not know until afterwards.
Three specific threats, three specific defences.
Phishing that reads perfectly. The tell-tale typos are gone. Do this: verify any payment or bank-detail change on a second channel before acting. Ring the number you already have on file, not the one in the email. Every time, no exceptions.
Voice-clone urgency calls. Scammers can now imitate the voice of a colleague, a supplier, or you. It always sounds urgent, because urgency is the point. Do this: set a call-back rule. No payment or account change happens on a single inbound call, ever.
Cloned websites and fake invoices. AI can copy a legitimate site’s branding at scale, right down to fabricated testimonials. Do this: type supplier web addresses yourself, and match ABN and bank details against your own records before paying anything.
None of this requires new software. It requires a written process your team actually follows. This is not a tool problem. It’s a systems problem.
The big picture: governments are now in the release room
The quieter story this week is who was standing next to the launch buttons. The White House reportedly asked OpenAI to delay its GPT-5.6 rollout, only weeks after Anthropic’s most advanced models were briefly taken offline under national security restrictions.
Frontier AI is starting to be treated like infrastructure, closer to banking than to apps. For a small business owner, I read that as good news. It means fewer chaotic surprise launches, more stable tools, and vendors who have to behave like utilities rather than startups.
It also means the sensible strategy has not changed: build workflows that survive model swaps. The model will change again by Christmas. Your workflow shouldn’t have to.
Bring it home
If this week has a theme, it is that capability is no longer the constraint. Prices are falling, agents are real, and the middle tier now does most of what a small business needs. The constraint is structure: knowing which of your workflows deserves an assistant, and which just needs fixing first.
That is the work I do with owners across Wollongong and the Illawarra. If you want a second set of eyes on where AI genuinely fits your operation, and where it does not, the AI and automation page explains how I approach it, or you can skip straight to a conversation.
See you next Monday. Structure, not AI noise.
Ashley
