你的 AI 只知道寫下來的內容;企業運作還依賴未被記錄的資訊
Your AI only knows what's written down. Your company runs on what isn't.
AI 智能體只能根據已記錄的內容工作,但企業決策也受員工掌握、難以寫下的背景脈絡影響。文章主張由智能體和員工組成混合智能團隊,透過雙向交接逐步記錄背景資訊。LeafCraft 正在打造讓智能體和員工協作的平台。
Last time I wrote about chess: two amateurs with a better process beat the grandmasters and the supercomputers. Today, the reason that process needs people at all.
The data is not the business
Every AI agent works from the same thing: what's written down. Tickets, emails, documents, databases, logs.
But think about the last important decision your company made. How much of it was in the data?
- The customer who said "it's fine" on the call, in a tone that meant it wasn't.
- The deal that's "pending legal", when the real blocker is a new CFO who doesn't trust you yet.
- The supplier who'll always squeeze in your order, because of a favour from three years ago.
- The machine on the plant floor that sounds wrong a day before the sensor notices.
- The reason a policy exists, which nobody wrote down because everyone who was there remembers.
None of that is in a system. All of it decides what happens next.
We know more than we can tell
The philosopher Michael Polanyi put it in one line in 1966: "We can know more than we can tell."
That's the problem in a sentence. Your people carry context they couldn't write down even if you asked them to. Not because they're hiding it, but because most of it only shows up in the moment: in a conversation, in a room, in a judgement call.
An agent, however capable, can't act on context it never sees.
So the answer isn't "more AI" or "fewer people"
If you hand everything to agents, they'll be fast, confident and missing half the picture.
If you keep everything with people, they'll drown, because the volume of work keeps growing.
The only model that works is hybrid intelligence: agents and people as one team, each doing what they're actually good at.
- Agents bring scale. They read everything, never forget what's written, and work around the clock.
- People bring context. They know the customer, the history, the politics and the moment.
- The work moves both ways. An agent knows when it's missing context and asks the right person, not "a human". The person answers, and that answer goes on the record, so the next time it's no longer missing.
Done well, every handoff makes the company a little smarter. The context that used to live only in people's heads slowly becomes something the whole team, people and agents, can work from.
What this changes for leaders
Stop asking "which jobs can AI replace?" and start asking a better question: "where does the context for this decision actually live?"
If it lives in the data, let an agent run it.
If it lives in a person, make that person a peer of the agent, not a step at the end of a queue.
If it lives in both, design the handoff.
That's what we're building at LeafCraft: one platform where AI agents and people work as one team, with clear jobs, handoffs both ways, and every decision on the record with its reason.
And that's what this newsletter is about. Subscribe if you're working through the same questions. I'll share what we're learning, including what doesn't work.
Where does the most important context in your company live today: in your systems, or in your people's heads? Tell me in the comments. I read and answer every one.
Where does your company stand? The 5-point check: https://leafcraft.ai/intelligent-enterprise/?utm_source=devto&utm_medium=crosspost&utm_campaign=hie20#check
Originally published as edition 2 of my LinkedIn newsletter, The Hybrid Enterprise. Edition 1: In 2005, two amateurs beat the grandmasters and the supercomputers. I'm the founder of LeafCraft, the Hybrid Intelligent Enterprise platform, and we build the LeafMesh ADK (pip install leafmesh).
Source: Michael Polanyi, The Tacit Dimension (1966).
來源:DEV Community · dev.to