Loading today’s stories
Product1 distinct publisher3 min readUpdated
A Fast Company essay describes an agent's email answered by a chatbot draft. With 51% of AI users at work using it to write, default behavior will set the norm unless someone writes one down.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
An AI agent wrote a courteous project status email, complete with next steps and named owners, and sent it on behalf of the person it represented; the recipient pasted it into his chatbot and sent back the reply the chatbot drafted, according to a Fast Company essay [1][2]. On the surface that exchange looks like two people corresponding, and underneath it is software corresponding with software [3].
The interesting part is not the novelty. It is that this happened before either side's employer had a rule about it. The essay argues that AI use forces organizations to confront questions their policies never anticipated: which conversations may be delegated, what a machine may commit the company to, and who owns what gets said [9].
Volume is what drives the drift. Microsoft's 2025 report on work trends found the average worker received 117 emails and 153 Teams messages a day [6], which is 270 inbound items before anyone writes a word [7]. Against that load, the essay argues, delegating to AI is increasingly the only rational response left to employees, and may even make them more productive [16]. Gallup reported last month that more than half of US employees now use AI at work, and the single most common use, cited by 51% of them, is writing and editing [5]. Multiply those two figures and roughly a quarter of the American workforce is already using AI to produce the words other people read [8]. Meanwhile Bloomberg in May profiled Tyler Cadwell, founder of an Arizona glassware business, who built an AI agent he calls his "first AI employee" and has it triaging his inbox and responding on its own to supply chain problems [4].
The essay's framework has three parts: decide what must never be delegated, make all other delegation deliberate, and rebuild ownership for the exchanges handed over entirely [10]. The first is the easiest to act on. Performance reviews, bad news, and mentoring derive their value from a human having actually done the work, so delegating them does not create efficiency, it destroys the thing the work was for [11].
The second is where most organizations will lose. Delegation runs along a spectrum: AI-assisted, where you create and the machine polishes; AI-delegated, where the machine drafts and you skim and send; and AI-represented, where your agent conducts the exchange and you may never see it [12]. An estate agent profiled by the Financial Times runs nine inboxes with an AI tool, saves hours a week, and admits that "sometimes I am guilty of letting it think for me" [13]. The essay's point is not that the tool thinks for her, which is what delegation is, but that no decision was ever made; chosen delegation comes with a handover, drifted delegation comes with none [14]. Across a workforce that becomes an organizational condition in which the company no longer knows what is being said in its name or how much human judgment sits behind it [15].
Note what the three pillars do not include: telling the recipient. Ownership and delegation are internal controls; disclosure is a norm between parties, and it is being set right now by whatever people happen to do.
Watch whether your own policy names a never-delegate list before an agent commits you to something, whether mail and CRM tools start stamping agent-sent messages by default, and what share Gallup reports for writing and editing next time it asks.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
A colleague received a courteous, professionally worded email reporting the state of a shared project, the next steps required and who was accountable for each; it was from start to finish the work of an AI agent set up to act on behalf of the person it represented.
The recipient of the agent-written email pasted it into his chatbot and sent the reply the chatbot drafted.
On the surface it looks like humans talking to humans; underneath, AI is talking to AI.
In May, Bloomberg profiled Tyler Cadwell, founder of an Arizona glassware business, who has built an AI agent he calls his "first AI employee"; the agent triages his email inbox and responds on its own to supply chain problems.
Gallup reported last month that more than half of U.S. employees now use artificial intelligence at work, and the single most common use, cited by 51% of them, is writing and editing.
Microsoft's 2025 report on work trends found that the average worker received 117 emails and 153 Teams messages a day.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single opinion source, third-party data cited secondhand
The cluster is one bylined essay. Its factual spine (Gallup usage shares, Microsoft 2025 message volumes, Bloomberg and FT profiles) is attributed but never linked or quantified beyond headline numbers, and no primary report is in the cluster. The load-bearing arguments about value destruction, organizational drift and productivity gains carry no measurement at all.
AI-assisted writing is mainstream; fully agent-represented exchange is anecdotal
Adoption splits sharply by rung of the essay's own spectrum. AI-assisted and AI-delegated writing is broad — a majority of U.S. employees using AI at work with writing and editing the top use, implying roughly a quarter of employees drafting or editing with AI. Fully AI-represented exchange, the story's actual subject, is evidenced by one firsthand anecdote and two profiled single-operator deployments, and the author concedes it is still unusual.
Framing runs ahead of the evidence, but the essay hedges
The premise that agent-to-agent email 'is already here' is carried by a single anecdote the author himself calls relatively unusual, and the organizational-drift and productivity consequences are asserted rather than measured. The overstatement is moderate rather than severe: the adjacent survey data on AI-assisted writing is real, the productivity claim is hedged as 'may even', and the piece explicitly recommends auditing to find out whether the condition exists.
Author's advisory and publishing business promoted in-line
The article body carries a promotional block for the author's books, podcast and companies, which 'give leaders the frameworks and platforms' for exactly the governance problem the essay defines. Publishing a named three-pillar framework and a four-move governance checklist directly advances that commercial offer, and no conflict note accompanies it. This is disclosed-by-presence rather than hidden, so it is a material but visible incentive.
Low-moderate: one publisher, no corroboration
One source, one publisher, no independent confirmation of the cited statistics or profiles, and an identifiable commercial incentive behind the framing. Confidence is highest for the descriptive framework and the existence of the cited datapoints, and low for the causal and organizational claims.
leadership
Microsoft puts AI agents in Entra, which makes agent sprawl an identity team problem1 distinct publisher
product
Nebius funds $4.5bn of AI capacity on terms that pay lenders mostly in stock2 distinct publishers
product
White House lets vetted firms hack back and leaves liability blank for 60 days1 distinct publisher
product
Pearson's inversion: in learning products, the correct answer can be the product failure1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 17, 2026