Dharmesh Shah — co-founder and CTO of HubSpot, writing to more than 2 million subscribers — just published a piece about scheduled AI tasks. His central argument: the biggest shift in how we use AI isn't about prompting better. It's about delegating and walking away. AI as asynchronous teammate, not search engine. I read it and thought: I already did this. This week. And the output is sitting right in front of me — because the blog posts you're reading now came directly from what three AI agents handed back to me on Tuesday morning.
What actually happened
I had a week's worth of newsletters stacked up. The kind of reading pile that used to eat a Monday morning — skimming, tabbing, trying to hold threads together across a dozen different sources while also answering messages and starting actual work. I've done that for years. It's not reading. It's triage.
This week I did something different. I briefed Claude CoWork — Anthropic's multi-agent system — on what I was looking for: the most significant AI and change management stories from the past week, the threads connecting them, and the angles most relevant to business leaders navigating AI adoption. One prompt. Then I walked away.
What I didn't expect was how it handled the task. Rather than processing everything sequentially, it split the job across three specialised agents running in parallel — one reading and extracting from the newsletters, one identifying cross-source themes and patterns, one synthesising everything into a structured brief with the angles most likely to resonate with my audience. I could watch the work happening in real time. Three separate threads of analysis, converging into a single output.
One prompt. Three agents running in parallel. One structured brief. Directly seeded this week's content.
What came back — and what I did with it
The brief that landed was genuinely useful. Not a summary of everything I'd missed — a curated, prioritised set of angles, with the connective tissue already drawn between stories I wouldn't have thought to link myself. The METR doubling curve research and the wemustactnow.ai economist statement were in there. The Gen Z sabotage data. The Micron chip factory. Each with a note on why it might resonate with the people I write for.
I read it in about eight minutes. Picked the angles I wanted. Briefed Claude Code to build the posts. You are reading the output.
What used to take a morning of fragmented reading and note-taking — the kind of cognitive work that feels productive but mostly just keeps you busy — took me less time than a cup of coffee to review. The human part wasn't eliminated. I still made the editorial calls: which angles had my voice, which stories had genuine change management relevance, what was interesting versus what was important. But the aggregation and pattern-recognition work — the part that doesn't require me specifically — was handled.
"The difference between a tool and a teammate." — Dharmesh Shah, Co-Founder & CTO, HubSpot
Why Dharmesh writing this matters
Dharmesh Shah is not an AI researcher. He's a builder of business software, and he writes for two million working professionals — people who use tools, not people who build them. When someone with that reach and that audience writes a piece explicitly framing AI as an asynchronous teammate, it signals that this shift has crossed a threshold. It's not early adopter territory anymore. It's mainstream business practice.
His examples are deliberately mundane: weekly competitive research reports, daily email triage, recurring market analysis. Not autonomous robot surgeons. Not AI-generated legislation. Just: here's the reading pile I used to do myself, and here's the colleague I now delegate it to. For $20 a month.
That price point and that framing are the story. The adoption barrier for this kind of workflow isn't technical ability or budget. It's imagination — the capacity to see your own work differently and ask which parts of it require you and which parts don't.
What this means for how teams actually work
Here's the change management angle that tends to get skipped in articles like Dharmesh's. When AI becomes an asynchronous teammate — when it does real work independently and hands you the output — it doesn't just change individual productivity. It changes how teams are structured, what a "role" means, and where accountability sits.
If one person with AI adoption fluency can do the synthesis work that used to require a team, what does that team do now? If the brief-writing, the research aggregation, the pattern-finding is handled by agents, what are the human skills that sit above that layer? These are not rhetorical questions. They're the structural questions that organisations need to be actively working through — and most aren't.
The practical adoption of multi-agent workflows is happening fast. The organisational thinking about what that means for roles, capabilities, and accountability is not keeping pace. That gap is exactly where change management lives.
The thing worth trying this week
Pick one recurring task in your week — the reading pile, the competitive scan, the stakeholder update, the meeting prep — and brief an AI to handle the aggregation. Not to write your response. Not to make your decisions. Just to do the gathering and pattern-finding that precedes your judgment.
See what comes back. Notice which parts you still need to do yourself. That gap between what the AI returns and what you add is a useful map of where your actual value sits. For most people, it's more precise — and more interesting — than they expected.
The adoption shift from tool to teammate doesn't happen all at once. It happens one delegated task at a time. Start with the reading pile.
