Lobbing a private Slack direct message to a colleague is second nature to many in the workplace-to brainstorm, plot strategy or even just vent. Artificial-intelligence creep is changing all that.

Executives are urging their staff to send messages in public Slack channels, open to anyone in the office to see, rather than privately pinging colleagues. They want AI agents crawling through messages to have unfettered access to every update on products, earnings and strategic initiatives.

As bots are unleashed throughout workplaces to analyze information, make recommendations, pull together slide decks or write code, the thinking goes, those AI agents should know the very latest information available. After all, information is power. And while humans control most of it, bots are faster at gobbling it up and sifting through it.

Much of the focus on AI's impact on work has highlighted jobs-how corporate ladders will be structured and how staff might be reduced. Within companies, however, many leaders are examining how AI will change internal norms to deepen the technology's impact. Communication is one of them: Data on what people are writing and how they are working are becoming extremely valuable to companies.

"A big part of getting these agents to work really, really well is to make sure they have as much context as you can feed it," said Wade Foster, CEO and co-founder of Zapier, which automates workflows for corporate customers. He tells staff to err on the side of public messaging: "You should be defaulting to doing this."

Foster said using public Slack channels has been encouraged for years, in part so that new hires could get up to speed. With AI agents, though, "that institutional memory becomes so much more valuable," he said.

Slack gives Zapier data on the percentage of messages that individuals send to public channels. Foster set up a friendly competition among executives this year: Whose percentage could be the highest? Foster recently clocked in at 79%; his chief marketing officer hit 98%. He said the goal isn't 100%, but could be around 80% for most leadership roles.

The practice extends to meetings. At Zapier, recorders save many virtual video conversations, tracking what was decided and discussed. Those notes are often uploaded to Slack channels.

More messages, many of which are designed primarily to be read by bots, will stay unread by colleagues. Reading it all, Foster said, "would not be a smart use of your time."

On a recent call with a customer, he asked a Cursor agent to build a slide presentation based on all Slack messages related to that customer over two months. It did so during the call.

No message sent at work is ever really private. Last week, a judge delayed Google's $10 million bid for digital archives from now-defunct Spirit Airlines that would include about 100 million emails, personnel records and Microsoft Teams messages. Google planned to use the data to improve its AI models, but the purchase was delayed after an objection from the union representing former flight attendants. Google maintains that any data it receives will be scrubbed of personally-identifiable information.

Employees should know that when agents are reading and acting upon their messages, they might also use that information for model-training purposes, said Mathijs De Vaan, an associate professor leading an AI strategic initiative at the University of California, Berkeley's Haas School of Business.

De Vaan said private messages offer employees a chance to brainstorm freely, which could be stifled with public messages.

Some students who are enrolled in part-time M.B.A. programs while working have told De Vaan that this practice occurs in their offices, he said. "The argument that there's important information exchanged in these DMs is correct," he said, and granting AI systems access to internal conversational data allows organizations to train agents more effectively.

Salesforce, which owns Slack, said channel usage and engagement among customers have grown recently. The company has rolled out a group-chat feature for multiple colleagues and agents to code together.

"Having more things in a system of record, having more things accessible, is the fuel that powers the innovation," said Joe Inzerillo, Salesforce's president of enterprise and AI technology. The same trend is playing out inside Salesforce, he added.

There are intuitive caveats, he noted-such as using a private message to chat with CEO Marc Benioff about an acquisition. "If there's not a reason for it to be private, then why wouldn't you put it in an open channel?"

Inzerillo said the method saves him significant time when he tries to recall where he heard a fact. He can get a quick answer by deploying an agent to check meeting transcripts that are shared on Slack.

The practice appears more common at AI labs or companies founded in recent years, after generative AI took hold in workplaces, Inzerillo said.

One such company is Aurelian, which sells software that coaches 911 dispatchers and automates nonemergency calls. Founder Max Keenan often asks colleagues how they can make information AI-accessible. He said he had "recently gone to war" against private workplace direct messages.

The cost of "information isolation" is too high, he added, and people and agents must know everything going on.

ElevenLabs, which creates AI-voice agents, only uses email for external communications. All internal messaging happens on Slack. At first, it was easy to keep up: The company started with 10 employees.

As the company grew to about 800 employees across 50 countries, it felt like it had reached the breaking point other companies hit when they outgrow a meeting room, said head of operations Victoria Weller.

Staff couldn't read every single Slack channel, she said.

Now, an agent pulls updates from the most relevant 100 public channels and turns them into a podcast that employees can listen to on their commutes.

When information isn't siloed, she said, "you don't need to find a person that was in the room for that conversation to get a readout of it."