The latest wave of AI news wasn’t about smarter chatbots. It was about AI quietly moving into the tools you already pay for — your Office software, your compliance stack, your document pile — while the price of using it dropped again. For an owner running a $1M–$20M business, the question is no longer “should we try AI,” it’s “which of these moves changes how my business runs this quarter.” Here are the five worth asking right now.
- Perplexity’s AI agent now works inside Word, Excel, and Outlook — what does that mean for my admin-heavy office?
- Anthropic just made HIPAA-compliant AI self-serve — should my med spa, dental, or clinic finally switch it on?
- AI prices dropped again — am I overpaying for what I actually use?
- Mistral’s new OCR can read almost any document — could it finally kill my paperwork pile?
- Google and Meta both stumbled this cycle — how do I avoid betting my business on the wrong AI vendor?
1. Perplexity’s AI agent now works inside Word, Excel, and Outlook — what does that mean for my admin-heavy office?
Perplexity pushed its “Computer” agent directly into Microsoft 365 — Word, Excel, PowerPoint, Outlook, and Teams — and paired it with a secure sandbox called SPACE that lets the agent actually do work without being handed the keys to your whole system. The company’s revenue crossed $450M and its valuation hit $22.6B, which tells you this isn’t a science project; it’s a product businesses are paying for.
Here’s why this is a bigger deal for real estate teams, insurance agencies, accounting firms, and other professional-services shops than for anyone else: your work already lives in Outlook and Excel. An agent that drafts the follow-up email, updates the spreadsheet, and preps the Teams recap without you copying and pasting between a chatbot and your real files removes the most annoying part of “using AI” — the shuffling. If you’re a real estate agent, that’s the difference between an assistant that suggests a listing description and one that writes it, drops it in the right doc, and queues the client email.
One caution came with the news: the SPACE sandbox exists because letting an agent write directly into your production systems is genuinely risky. Let it draft and prepare, but keep a human clicking “send” and “save” until you trust it — and start with one repetitive workflow, not your whole back office.
2. Anthropic just made HIPAA-compliant AI self-serve — should my med spa, dental, or clinic finally switch it on?
Anthropic made two moves that matter if you handle patient information: it turned on self-serve HIPAA support (meaning you can sign the compliance paperwork and start using its Claude models without a lengthy enterprise sales process), and it made its faster, cheaper Sonnet 5 model the default. It also extended free access to its Fable 5 model again, so the cost of testing is basically zero.
For a med spa, dental practice, physical therapy clinic, or small medical office, the HIPAA piece is the unlock. Until now, the honest answer to “can I use AI on anything involving a patient” was usually “not without legal risk.” Self-serve HIPAA changes that math for intake summaries, appointment reminders, insurance-question triage, and post-visit follow-ups — the front-desk work that eats your staff’s day. The compliance foundation being self-serve means a 6-person clinic can now set this up the same way a hospital would, without a hospital’s IT department.
This isn’t a green light to feed protected health information into any random chatbot — the compliance coverage only applies when you’ve actually signed the agreement and configured it correctly. But if you’ve been sitting on the sidelines specifically because of privacy rules, that specific excuse just got a lot weaker.
3. AI prices dropped again — am I overpaying for what I actually use?
Three separate things happened at once here. OpenAI split its new GPT-5.6 into three tiers — Sol (the flagship), Terra, and Luna — where the mid-tier roughly halves the cost of high-volume tasks at near-flagship quality. xAI released Grok 4.5 at roughly $2 to $6 per million tokens, undercutting comparable top-tier models by around 4x. And DeepSeek rolled out cheaper off-peak pricing on its open-weight V4 model. The through-line: the “one model for everything” approach is now the expensive way to do it.
The practical version of this question is for any business running AI at volume — e-commerce stores handling product descriptions and support tickets, retailers doing customer replies, agencies generating drafts all day. You almost certainly don’t need the most powerful (and most expensive) model for every task. Routing your high-volume, low-complexity work to a cheaper tier and saving the flagship for the hard stuff is where real savings live. If you’re paying a flat premium rate for everything, you’re leaving money on the table.
You don’t need to understand the engineering to act on this. Ask whoever manages your AI tools one question: “Are we using the cheapest model that still does the job for each task?” A blank stare is your audit finding right there.
4. Mistral’s new OCR can read almost any document — could it finally kill my paperwork pile?
Mistral shipped OCR 4, a major upgrade to its document-reading technology, alongside open models you can run on your own hardware. OCR — turning scanned paper, PDFs, and photos into usable digital text — sounds boring until you realize how much of your week is spent re-typing information off documents that a machine could now read in seconds.
This one is written for the paperwork-heavy trades and back offices: contractors and home-services companies processing permits, invoices, and supplier quotes; property managers handling leases and maintenance receipts; insurance shops keying in claims; medical billers working through statements. If your team is manually pulling numbers off documents into a spreadsheet or your job-management software, modern OCR can do the first pass and hand your staff a review-and-approve task instead of a data-entry task. The added twist is that Mistral’s models can run on your own hardware, which matters if you’d rather not send sensitive documents to an outside cloud at all.
The realistic starting point is one document type you handle constantly — say, supplier invoices — and measuring how many minutes per document you save. That number, multiplied by your monthly volume, is usually what turns “interesting” into “obvious.”
5. Google and Meta both stumbled this cycle — how do I avoid betting my business on the wrong AI vendor?
Not everything went smoothly. Google’s Gemini 3.5 Pro missed its release target for the third time and reportedly got rebuilt from scratch amid a talent exodus. Meta shipped some tools but pulled others and saw its splashy “superintelligence” push fall flat. Meanwhile Cohere agreed to acquire Germany’s Aleph Alpha in a roughly $20B combination aimed at businesses that need their data to stay in a specific country. The lesson isn’t “these companies are doomed” — it’s that even the giants slip, and their roadmaps are not promises.
For a business owner, this is the vendor-lock-in question, and it applies no matter your industry. If you build a core workflow around one company’s model and that company delays, changes its pricing, or retires the version you depend on, you inherit the disruption. The owners who stay calm through these cycles are the ones who kept their setup flexible — using tools that can swap the underlying AI model without rebuilding everything.
You don’t need to become a technology strategist. You just need to ask, before you commit: “If this vendor changed their pricing or discontinued this tool next year, how hard would it be for us to switch?” If the answer is “very hard,” build in a fallback now, while it’s cheap to do so.
What This All Means
The headline this cycle is that AI stopped being a separate destination and started showing up inside the software, compliance rules, and documents you already deal with — and it got cheaper while doing it. The businesses that win won’t be the ones chasing every announcement; they’ll be the ones who pick one or two shifts that map to their actual daily work and act on them. Start with a single repetitive workflow, measure the time saved, and expand from there. The technology finally caught up to the boring, expensive parts of running a business — which is exactly where the money is.
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 →