The AI headlines this week weren’t really about who built the smartest chatbot. They were about three things every business owner actually cares about: what AI costs, what you’re allowed to use it for, and whether it can finally do the work instead of just talking about it. OpenAI, Anthropic, xAI, Perplexity and a wave of European challengers all moved on those exact fronts. Here are the five questions to ask before your next AI decision — several of them written for a specific kind of business.
- Am I overpaying for AI on tasks that don’t need the smartest model?
- What does Perplexity’s new Microsoft 365 agent mean for real estate agents and bookkeepers specifically?
- Can medical spas and dental practices finally use AI on patient data without breaking the rules?
- Should I lock into one AI vendor while prices are collapsing?
- Are AI “agents” actually ready to run part of my operation now?
1. Am I overpaying for AI on tasks that don’t need the smartest model?
OpenAI released GPT-5.6 this week in three tiers — Sol (the flagship), Terra (mid-range), and Luna (lightweight) — alongside a new “ChatGPT Work” product. The detail that matters for your budget: Terra delivers roughly the quality of last generation’s premium model at about half the price. Translation — a lot of the AI work you’re paying top dollar for right now can be done for materially less.
This is a bigger deal than it sounds if your business runs high volume. Think of a home-services company generating hundreds of appointment confirmations, follow-up texts, and review requests a week, or a retailer writing thousands of product blurbs. You don’t need the most expensive model to draft a “we’re running 15 minutes late” text. You need the cheap tier for the routine 90% and the smart tier for the 10% that’s actually hard. The industry term is “task-tiered routing,” and the practical version is simply: stop paying flagship prices for grunt work.
If you’re on a flat AI subscription and have no idea which tier is doing your work, that’s the question to bring to whoever set it up. With xAI’s Grok and China’s DeepSeek both undercutting the majors on price this same week, the cost of routine AI is dropping fast — and you should be capturing that, not the vendor.
2. What does Perplexity’s new Microsoft 365 agent mean for real estate agents and bookkeepers specifically?
Perplexity pushed its “Computer” agent directly into Microsoft 365 this week — meaning it can now act inside Word, Excel, PowerPoint, Outlook, and Teams, not in a separate chat window you copy-paste from. It also launched a secured sandbox called SPACE so the agent can work at full capability without loose access to your files.
If your business lives in Office, this is aimed squarely at you. A real estate agent spends hours turning listing notes into descriptions, drafting buyer follow-ups in Outlook, and updating comp spreadsheets — all inside the exact apps this agent now operates. A bookkeeper or small accounting firm lives in Excel and Outlook. An agent that can open the spreadsheet, reconcile the rows, draft the client email, and leave it in your outbox for approval is a different animal than a chatbot you have to babysit.
The honest caveat: “acts inside your files” is powerful and slightly scary. That’s why the sandbox announcement matters — you want an agent that works in a controlled space and hands you finished drafts to approve, not one with unsupervised write access to your live systems. The question for your business is which repetitive, in-Office task you’d hand over first, and what approval step you’d keep in the middle.
3. Can medical spas and dental practices finally use AI on patient data without breaking the rules?
Anthropic quietly shipped something this week that regulated small businesses have been waiting on: self-serve HIPAA support for Claude. Previously, using AI on anything touching patient information meant lawyers, enterprise contracts, and a compliance headache most small practices couldn’t justify. Self-serve means a med spa, dental office, or therapy practice can set up compliant use without an enterprise sales process.
Why this is the week’s most underrated story for local health businesses: your bottleneck was never whether AI could draft the post-treatment follow-up, summarize the intake form, or answer the “is this normal after my filler?” message. It was whether you were allowed to feed patient information into it. That barrier just got dramatically lower.
Anthropic paired this with public commitments around data handling and even a free program for K-12 teachers — a signal that “trust and compliance” is now a competitive battleground, not an afterthought. For any owner in healthcare, dental, aesthetics, or wellness, the question is no longer “can I use AI?” It’s “which patient-facing workflow do I make compliant first — and who on my team owns the approval step?”
4. Should I lock into one AI vendor while prices are collapsing?
This week made the case for keeping your options open better than any advisor could. In seven days: OpenAI cut effective prices with its Terra tier, xAI’s Grok 4.5 was reported to undercut a leading premium model by roughly 4x, DeepSeek shipped a powerful open-weight model with peak/off-peak pricing, and Anthropic extended free access to its Fable 5 model for a third time. Meanwhile Google’s much-anticipated Gemini 3.5 Pro missed its target date for the third time and is being rebuilt.
The lesson for a $1M–$20M business is not to pick a “winner.” It’s to build so you can switch. The labs are competing between launches now, and prices and quality are moving month to month. If your operations are hard-wired to a single provider’s specific product, you can’t take advantage of a competitor halving its price next month — and you’re exposed if your one vendor slips a deadline like Google just did.
Practically: whatever AI tool you adopt, ask your vendor or consultant how hard it would be to swap the underlying model if a cheaper or better one appears. “Model-swappability” should be a buying criterion, not an afterthought. One caution worth flagging — DeepSeek is retiring some older endpoints and is a Chinese provider, so if data sovereignty matters in your industry, factor that in before chasing the lowest sticker price.
5. Are AI “agents” actually ready to run part of my operation now?
The word “agent” got thrown around all year, but this week showed the shift from demo to deliverable. OpenAI’s ChatGPT Work is built to hand back finished outputs, not suggestions. Perplexity’s SPACE gives agents a secure place to actually execute. Mistral launched “Vibe,” an agent with a dedicated Work Mode and Code Mode. The common thread: agents that complete a task end-to-end and hand it to you, rather than chatting about it.
For a service business — HVAC, plumbing, landscaping, a clinic, an agency — the realistic first use isn’t replacing your team. It’s taking one annoying, repeatable workflow off their plate: turning a completed job into an invoice and a review request, drafting the weekly schedule from incoming bookings, or chasing overdue payments. The technology crossed a line this week where “the AI does the whole task and I approve it” is genuinely on the table for small operators, not just enterprises.
The right question isn’t “should I get an agent?” It’s “what is the single most repetitive task in my week that I’d happily let a machine draft and me approve?” Start there, keep a human approval step, and expand only once it earns trust.
What this all means
Strip away the model names and this week tells one story: AI is getting cheaper, more trustworthy for regulated work, and finally capable of finishing tasks instead of just discussing them. That’s a good thing for small and mid-sized businesses — the barriers that used to favor big companies (enterprise contracts, huge budgets, technical teams) are all coming down at once. The owners who win won’t be the ones who chase every launch; they’ll be the ones who pick one high-value workflow, keep their options open on vendors, and put a human in the approval seat.
Not sure how any of this applies to your specific business? That’s exactly what our free audit is designed to answer. In 30 minutes, we’ll map out where AI can realistically save you time, generate leads, or give you visibility you don’t currently have — no jargon, no pressure. Book your free audit →