AI moves into personal channels — more time for small business, more risk too
Stampli's team cut two-thirds of their hours, ChatGPT starts writing SMS, and a Grok security flaw shows why caution matters.
Summary
- Stampli’s marketing team finished 243 hours of work in 77 — using ChatGPT Work. A concrete number that a 3-5 person office can replicate.
- ChatGPT starts drafting and sending SMS through Apple Messages — for a small business owner, that means customer replies and follow-ups can be AI-drafted; you just approve.
- A serious flaw in Grok — the lesson for every AI user: never ask a chatbot to summarise a page from a link you don’t trust.
Big picture: AI is no longer “a separate website where you go and ask.” It’s moving into phone messages, email, everyday tools — which brings real time savings, but also new rules you can’t afford to ignore.
1. A small team finished 243 hours of work in 77
What changed. OpenAI published the Stampli case study — the marketing team at the accounts-payable automation company used ChatGPT Work and Codex together to compress a product-launch cycle that normally takes ~243 hours down to ~77. That’s a 68% time saving. Hundreds of content pieces per week now, versus a handful before.
What it means for a small business. This isn’t a story about a big company with an AI team — Stampli’s marketing team is small by description, and the method is simple: humans still approve everything that reaches customers, but drafts, research and repetitive pieces are done by AI. If you run a 3-5 person business and personally write the social posts, customer proposals, blog posts — the same method works for you. Prerequisite: one repeating task per week that takes more than an hour.
What you can do this week.
- Pick one task you do every week for ≥ 1 hour (customer replies, monthly report, social posts, quotes). Time yourself doing it once.
- Next time, do it with ChatGPT Plus or Claude Pro (€20/mo) — give the tool prior examples and ask for a first draft. You edit and approve.
- Write down both times and the saving. If it’s 30% or more, repeat it next month. If less — try a different task.
What to watch.
- More concrete savings numbers from small businesses (not just tech companies) will appear in the coming months.
- ChatGPT Business/Team plans (€25-30 per user/mo) will become more accessible for smaller teams — they add team memory and shared documents.
2. ChatGPT enters Apple Messages — less distance between you and the customer
What changed. TechCrunch reports that a new ChatGPT plug-in for Apple Messages lets you sort, search, draft and even auto-send SMS. The plug-in runs locally on the computer and doesn’t build a full message index. Available in both personal ChatGPT and ChatGPT Work.
What it means for a small business. If you run a small service business — hair salon, auto repair, home services — and customer communication happens over SMS or iMessage, this is the first real step where AI can draft replies to “what time?”, “any slot left?” style messages for you. But be careful: OpenAI itself warns against enabling autonomous send — one bad text to a customer costs more than the seconds you saved.
What you can do this week.
- If you already have ChatGPT Plus (€20/mo) and a Mac — install the new Messages plug-in and for one week try “draft a reply”, but SEND YOURSELF. Don’t rush into autonomous sending.
- If customers write to you on WhatsApp Business or just to a phone number — stay with the smaller step for now: copy-paste the question into ChatGPT, draft the reply, send manually. That alone saves 30-60 seconds per message.
- Write down 5-10 typical customer questions and your standard reply style in one document — give it to the tool as “this is how I write, this is my business.” The tone will be right from the first try.
What to watch.
- Google Messages and WhatsApp Business will add similar AI features soon — expect announcements in the next 2-3 months.
- Data-protection questions (customer data handling, GDPR) will need thinking through — if a customer sends a private question, AI sees it.
3. A security flaw in Grok — a lesson for every AI chat user
What changed. Adversa AI security researchers disclosed a technique that can trick Elon Musk’s Grok chatbot: if you ask it to summarise a certain web page, encrypted commands hidden in that page can make Grok send the attacker your name, location, subscription tier, and even chat history. Grok received the report in June but has not shipped a fix yet.
What it means for a small business. This is not just a Grok problem — it’s a lesson about how AI chats work in principle. If you use ChatGPT, Claude, Grok or Gemini in your business (for handling customer data, summarising documents, writing emails), know that everything you show it or ask it to open is potentially visible. The specific encrypted attack is rare, but simpler risks (sensitive info stays in model history, customer data goes to US servers) are everyday.
What you can do this week.
- Never ask a chatbot to “summarise this page” for an unknown link a customer sent you or that came from social media. Open it yourself first.
- If you use ChatGPT or Claude for customer data — turn on privacy settings that prevent your data from being used for training. ChatGPT: Settings → Data Controls → turn off “Improve the model for everyone”. Claude: off by default.
- Don’t put personal data (national ID, bank details, health info) into a regular chat at all — if you must process it, use ChatGPT Team/Enterprise or local solutions.
What to watch.
- Latvia’s Data Protection Authority (DVI) plans to publish guidance on generative AI in small business by autumn 2026 — worth following.
- Every major chat service will offer a “zero data retention” (data not stored) version this year — OpenAI already released this for frontier models.
Today’s picture
Three stories point to one structural shift: AI is no longer a separate destination. A year ago, AI meant “go to chatgpt.com and type a question.” Today AI is moving into phone messages, office tools, marketing processes. Each such step means real time savings for a small business — but also new assumptions you can’t take for granted, namely that the tool always acts in your interest.
| Event | Consequence |
|---|---|
| Stampli’s 68% time saving with ChatGPT Work | Small teams can realistically perform like 3x larger ones |
| ChatGPT enters Apple Messages | Customer comms get faster, but you need strict “approve yourself” discipline |
| Grok security flaw | Every AI user should know that everything visible is potentially outbound |
Three questions for a small business owner:
- What’s the one repeating task in your week to start an AI experiment with?
- Do your employees know what data can and cannot be put into AI chats?
- When a customer sends you a link with a question, how do you handle it now — and will that change?
Sources
- OpenAI — Stampli case study: https://openai.com/index/stampli/
- TechCrunch — ChatGPT Apple Messages plug-in: https://techcrunch.com/2026/08/20/chatgpt-can-now-send-texts-for-you-with-new-apple-messages-plugin/
- The Hacker News — Grok Cryptographic Context Injection: https://thehackernews.com/2026/08/new-cryptographic-context-injection.html
- Ars Technica — Grok exfiltrates user data: https://arstechnica.com/security/2026/08/grok-exfiltrates-user-data-when-malicious-instructions-are-encrypted/
- OpenAI — Zero Data Retention: https://openai.com/index/offering-zero-data-retention-for-frontier-models/
AI news for business leaders
Twice a week: what AI news means for a small business. No hype, with concrete steps.