Latest / Future of Work Tech with Fexingo: Remote Tools, AI Productivity, and Workplace Software / How AI Is Ghostwriting Your Internal Chat Messages
Transcript
- Lucas: So you're chatting with a colleague on Slack about the Q3 forecast, and you type "Hey, got the numbers — looks like we're tracking about 4 percent above plan." You pause, then backspace the whole thing and let the AI suggest a message instead. Luna: I've done that. More often than I'd like to admit. Lucas: Right. And you're not alone. A survey from earlier this year found that 55 percent of knowledge workers now use AI to draft at least some of their internal messages. Not external — internal chat, the stuff you'd normally just type out on the fly. Luna: That's a big number. But I wonder — does it actually save time, or does it just shift where you spend the time? Because you still have to review and edit the AI's suggestion. Lucas: That's exactly the question we're going to drill into today. I want to look at one specific deployment — a mid-size tech company, about 800 employees, that rolled out an AI message-drafting feature for their customer support team inside Slack. They tracked everything. Luna: Okay, what did they find? Lucas: Average response time dropped by 40 percent. But here's the interesting part — the biggest time savings weren't on the simple messages like "Got it, thanks." It was on the complex ones. The replies that require context — referencing a previous conversation, pulling in a ticket number, adjusting tone. Luna: Huh. So the AI is actually better at the stuff we think of as uniquely human? Lucas: In some ways, yes. Because these tools are trained on the company's own message history. They learn the specific shorthand, the inside jokes, the way your team says "let's circle back" versus "let's regroup." Luna: But that also means they're learning your bad habits. If your team communicates in passive-aggressive corporate speak, the AI will reproduce that perfectly. Lucas: That's a real risk. The company I mentioned actually had to retrain the model after two weeks because it started mimicking a manager who had a tendency to be overly terse. New hires thought the AI was being rude to them. Luna: Yikes. So there's a governance piece here — someone has to audit what the AI is learning. Lucas: Exactly. And that's the part most vendors don't emphasize in their demos. They show you the slick interface, the time saved. They don't show you the meeting where you have to explain why your chatbot started writing like that one guy in accounting. Luna: I want to come back to that governance angle. But first — how does the AI actually decide what to suggest? Is it just predicting the next word, or is it doing something more sophisticated? Lucas: It's a mix. The underlying model is a large language model fine-tuned on the company's message corpus. But there's also a rules layer on top — certain messages get flagged if they contain sensitive keywords, or if the tone analysis scores above a certain threshold for negative sentiment. So the AI will say, "I can draft this, but it sounds frustrated — do you want to rephrase?" Luna: That's actually kind of helpful. I've definitely sent messages I regretted because I was in a bad mood. If the AI catches that, maybe it's worth the trade-off. Lucas: And that's where the efficiency versus authenticity debate gets interesting. Because sometimes the slightly rough, human message is better. It signals that you're actually engaged, not just auto-piloting. Luna: There's a psychological layer too. If I know you're using AI to write your messages, I start to wonder — is this really you? Or is this the average of everyone you've ever chatted with? Lucas: A team at Stanford actually studied that. They found that when people knew a message was ai generated, they rated the sender as less trustworthy — even if the message itself was perfectly fine. The mere suspicion eroded the relationship. Luna: So we're in this weird middle ground where the technology works, but the social norms haven't caught up. Nobody puts a disclaimer on their Slack message saying "drafted with AI." But maybe they should. Lucas: Some companies are experimenting with that. A few startups now append a small icon — like a little sparkle emoji — next to messages that were ai assisted. It's opt-in, but early data suggests it actually increases trust, because people feel like you're being transparent. Luna: That is fascinating. Let's talk about who's actually adopting this. Is it just tech companies, or are we seeing it spread? Lucas: It's spreading fast. The survey I mentioned earlier — the 55 percent number — that covered industries from healthcare to manufacturing. The biggest adoption is in customer-facing roles, but internal comms teams are also using it to draft company-wide announcements. Luna: That feels like a slippery slope. If the CEO's all-hands message is written by AI, you lose that personal connection. Lucas: But what if the CEO is a terrible writer? I've seen all-hands emails that were so convoluted nobody understood them. An AI that clarifies and tightens the language could actually improve communication. Luna: True. But there's a difference between editing and generating. If the AI starts from scratch, you're essentially outsourcing the CEO's voice. Lucas: Right. And that's where the line gets blurry. Most tools let you choose the level of assistance — from just correcting typos to full composition. The company I studied set the default to "suggestions only," meaning the AI would offer a draft, but the human had to actively select it. They found that within a month, 70 percent of messages were still being typed from scratch. Luna: So people are using it selectively. That's probably the sweet spot. Lucas: It is. The key is giving people the option, not forcing it. And that's where the product design matters a lot. The best implementations make the AI feel like a collaborator, not a replacement. Luna: You know, this conversation reminds me of something — a couple of dollars a month is genuinely what keeps shows like this going. If today's tech conversation gave you something usable, consider tossing a few bucks at buy me a coffee dot com slash fexingo. It keeps us ad-free and independent. Lucas: Yeah, exactly. Listener support is what lets us dig into these nuanced angles without worrying about sponsors. So if you've gotten value, it really helps. Luna: Okay, back to the AI messaging. I want to talk about the future — where is this heading in the next year or two? Lucas: The next frontier is multimodal. Imagine an AI that not only drafts your text but also suggests a relevant chart or screenshot from your desktop. Or one that can summarize a long thread and propose a reply that references data from a spreadsheet you have open. Luna: That would be incredibly powerful. But also a privacy nightmare. The AI would need access to everything on your screen. Lucas: Absolutely. And that's the biggest barrier right now. Enterprises are hesitant to give AI that level of access, even if it's on-device. Microsoft and Google are both working on on-device models that process everything locally, but the performance trade-offs are significant. Luna: So we're in this phase where the technology is possible, but the trust infrastructure isn't there yet. Lucas: Exactly. And that's why I think the companies that win in this space won't be the ones with the best model — they'll be the ones with the best privacy guarantees. The ones that can prove your data never leaves your device. Luna: Let's pivot to the human side. How do employees feel about this? Do they see it as a helpful assistant or a surveillance tool? Lucas: It depends entirely on how it's introduced. In companies where leadership framed it as "we're giving you a tool to reduce busywork," adoption was high and sentiment was positive. In companies where it was rolled out as "we're tracking message effectiveness to improve productivity," people freaked out. Luna: So it's all about framing. That's a lesson that applies to any AI deployment. Lucas: It really is. And the companies that got it right invested heavily in change management — town halls, pilot groups, feedback loops. They treated it as a culture change, not a software update. Luna: Last thing — what's the one takeaway you want our listeners to remember? Lucas: That AI ghostwriting is already here, and it's only going to become more pervasive. But the choice of how to use it — whether to let it smooth over your rough edges or to let it speak for you entirely — that's still a human decision. And it's one worth making deliberately. Luna: Well said. Thanks, Lucas. Lucas: Thanks, Luna. Talk to you next time.