Tool of the Week: AI teammates that live in your chat (Claude Tag)
Last week the busywork pitch came from Microsoft. This week it's Anthropic's turn, and the pattern is now impossible to miss.
Anthropic launched Claude Tag — an "always-on" Claude that lives inside Slack. You @Claude in a channel and it answers, takes on tasks, breaks them into steps, uses your connected tools, and replies right in the thread. Three things make it different from the chatbot you already know:
It remembers. It follows the channel over time and builds up context about how your team actually works — not just the message in front of it.
It's shared. There's one Claude identity per channel, so anyone can see what it's working on and pick up where a coworker left off.
It's proactive. An "ambient" mode posts updates on its own, surfaces relevant info from other channels, and chases the threads everyone forgot about.
It's in beta for Claude Enterprise and Team plans, so most small shops can't switch it on yet. But this isn't a demo: Anthropic says it now routes roughly 65% of its own code changes through an internal version of the tool. You don't need it installed to learn the lesson it's teaching.
Here's the part the headlines skipped. A tool that "follows along and learns ever more about the work" isn't only a productivity feature — it's a memory feature. Every decision, every why-we-do-it-this-way, every undocumented process gets absorbed into a system you don't own. That's genuinely useful. It's also a quiet switching cost: the longer it runs, the more of your company's institutional knowledge lives inside one vendor's product instead of yours.
I run my own team of AI agents day to day, so I'm not anti-this — I lean on exactly this kind of setup. But I keep one rule: the context belongs to me, the AI is just renting it. The knowledge lives in docs and systems I control, and the AI reads from them. Not the other way around.
What to actually do — whether or not you're on Slack Enterprise:
Write down the stuff that only lives in your head. Your refund policy, your "good lead vs. bad lead" rules, how you price a rush job. If an AI can learn it by watching, you can write it down once and own it.
Keep your knowledge in a place you control — a Google Doc, a Notion page, a folder — and let the AI read from that. Then you can switch tools without losing your memory.
Turn the proactive features on slowly. "Always-on and posting on its own" is powerful and easy to over-trust. Let it observe before it acts, same as any new hire.
Who this is for: Any owner watching "AI coworkers" land in Microsoft, Anthropic, and Google products and wondering what it means for them. The tools are real and getting better fast. Just make sure that as they learn your business, you still own what they learned.
Quick Hits
Companies are cooling on "AI spend for its own sake." Enterprise AI budgets averaged roughly $11.6M in 2026, yet about 56% of CEOs report no measurable revenue or cost benefit — and investors are now mocking "tokenmaxxing," the habit of inflating token-usage numbers to look productive. Why it matters: the tide is turning toward provable outcomes, and that's the small-business advantage. Pick one task, automate it, measure the hours saved. Outcomes beat usage — every time.
Cheaper, capable AI is coming out of China. A Chinese lab, Z.ai (Zhipu), released an open-weight model, GLM-5.2, that engineers say rivals top US models at a fraction of the price — and it's drawing real attention in Silicon Valley. Why it matters: more capable low-cost models mean downward price pressure on the AI subscriptions you already pay for. You may never touch a Chinese model, but you'll likely benefit from cheaper options on the tools you do use.
Prompt of the Week: The Institutional Knowledge Extractor
The theme this week is owning your context. So instead of letting an AI quietly learn how your business runs, spend 15 minutes pulling it out of your head into a doc you keep. Paste this:
Act as an operations analyst. Interview me to capture the undocumented
knowledge that keeps my business running — the stuff that only lives in my head.
Ask me one question at a time, then go deeper based on my answer. Cover:
- How I decide what's a good customer vs. one to turn down
- How I price, including exceptions and rush jobs
- My step-by-step for the 3 things I do most often
- The mistakes a new hire always makes, and how to avoid them
- Who I call when something specific goes wrong
When we're done, organize everything into a clean "How We Operate" document
I can save, reuse, and hand to a person or an AI later.Run it once and you'll have a document that makes hiring easier, onboarding faster, and — not incidentally — means your business's memory lives somewhere you control, no matter which AI tool you plug in next.
Like what you're reading? Forward it to someone who'd get value from it. And if you're curious what AI could actually do inside your business, book a free 15-minute audit — no pitch, just a look at where you're leaving time on the table.