Sunday night I asked my Clawdbot (now moltbot??) to brief me on Monday's schedule. I sent an audio message and expected a voice note in return so I could listen to while making coffee. It pulled my calendar, caught a conflict I'd missed, and dropped an MP3 in my Slack and Took about 30 seconds.
I also have a human VA that could have done that but it’s slower, and the UX is worse.
I've used virtual assistants on and off for almost a decade always hoping that for $2,000-3,000 a month, a dedicated human can handle the admin work that shouldn't require my attention. Calendar, travel, inbox triage, research. All these tasks follow a pattern, can be documented in an SOP, and executed over and over with minor variations.
The economics make sense. An executive assistant in Denver or San Francisco costs $80-120k fully loaded. A VA from the Philippines costs $24-36k a year. The generally do the same tasks, just at a lower price. The arbitrage was labor cost, and for years that arbitrage worked.
The Training Problem
I always get frustrated with a new VA the first two to three months are a net loss.
I’m recording Loom videos. Writing documentation. Answering questions. Correcting mistakes. Re-explaining preferences. Essentially compiling myself into a format another human can execute.
But humans are lossy. They forget. They interpret. They have off days.
I've done this cycle a few times. Each time I told myself the ROI will kick in eventually. Sometimes it did. Sometimes it didn’t.
Enter Clawdbot (or moltbot...)
Last week I set up Clawdbot, an open-source AI assistant that runs locally and connects to the services I use. Calendar, email, Notion, Slack. The setup took an afternoon, and was easy considering I am happy in a CLI. No training videos. No SOPs. No onboarding calls.
Monday I needed competitive research on Focused's market position. Normally this is a half-day project for a VA: Google around, pull together a document, present findings, field follow-up questions, iterate. I told Clawdbot to research our competitive landscape and come back with findings. About a half hour later I had a structured analysis covering partnerships, competitors, market sizing, and positioning. With sources cited.
My VAs have been smart, competent people. But AI assistants can use tools directly, reason about novel problems without explicit training, and don't require me to have thought through the exact steps in advance.
I never taught Clawdbot how to research podcasts. I never documented my process for finding conferences. I just asked, and it figured it out.
The Token Factory
Throughout my career I've watched companies outsource administrative functions to lower-cost labor markets. Call centers to India. Customer support to the Philippines. Virtual assistants to Latin America. Each move was the same arbitrage: American labor is expensive, overseas labor is cheap, the tasks are trainable, and the margin pays the coordination overhead.
The VA industry exists because humans in Manila can produce administrative output, emails sent, meetings scheduled, research compiled, at $15-25 per hour instead of $50-75. They are, in the language of AI, token factories. Trained on specific procedures, producing output according to that training, measured by throughput.
Now a cheaper token factory exists.
My Clawdbot instance costs roughly $200 a month in API calls. It runs 24 hours a day. No holidays. No timezone coordination. No departure for a better opportunity.
I use Athena, one of the larger VA services. They charge $3,000 a month for a dedicated assistant. That's $36,000 a year for someone who works my timezone, speaks fluent English, and handles my tasks. Compared to a $100,000 executive assistant, it’s a good deal.
Compared to $2,400 a year for an AI that never sleeps and improves automatically? The economics aren't close. But the quality is similar.
What Remains
I don’t think AI replaces all human work. There are tasks I'd never delegate to Clawdbot. But when I audit the tasks I've historically given to VAs, the majority are pattern matching and execution. Inbox triage is classification. Calendar management is constraint optimization. Travel booking is search and filter. Research is exactly what these models were built for.
The premium VA services will survive by moving upmarket—less like task executors, and become chiefs of staff. The commodity layer, the "handle my calendar and book my flights" tier, is dead. The economics have shifted, and economics always win eventually.
What Comes Next
I have a goal to speak at six conferences this year and appear on two podcasts a month. In previous years this would've been a project. I’d Research opportunities, track deadlines, draft pitches, follow up on responses. Multiple VA hours over multiple weeks, plus a spreadsheet I'd need to maintain.
I told Clawdbot to build me a pipeline. Find relevant podcasts, identify conferences with open CFPs, organize by deadline, and schedule time on my calendar to write my pitches. It's working on that right now as I write this.
It feels like science fiction, AI doing work that used to require human coordination. The VA category made sense when labor arbitrage was the only way to reduce administrative costs. Now there's another option, and that option is cheaper, faster, and improving at the rate of AI progress.
The virtual assistant is dead. And honestly? I think that's a win, unless you're a virtual assistant. I don't have a clean answer for that part yet, but I'm not going to pretend the economics are going to wait for one.
@austinbv is the CEO of @focused_dot_io, an AI consultancy that builds agents which actually work in production.*