Eighteen months ago I had three virtual assistants. One did my calendar and inbox triage. One did client reporting and social scheduling. One did lead follow-up and CRM cleanup. Good people, all three. Reliable, careful, did what I asked.
I don't have any of them anymore. Not because they were bad at the job. Because the job changed shape underneath them, and I built a system that does what they did, plus a fair bit they never had time for, and it never sleeps, never takes annual leave, and never needs me to re-explain a process for the fourth time.
So let's be clear about what I actually mean by "replaced", because it's not as clean as the headline makes it sound.
I want to walk you through exactly what changed, what I kept, what I lost, and where I think this is actually heading for people running a small team. Because there's a lot of noise about AI "replacing jobs" and most of it is either doom or hype. This is neither. This is just what happened when I actually did it.
The three jobs, honestly described
Calendar and inbox VA: triaged my emails into folders, booked meetings, chased people who hadn't replied, sent me a daily summary. Maybe 12 hours a week.
Reporting and social VA: pulled numbers from a handful of dashboards into a template every Monday, scheduled the week's LinkedIn and Facebook posts from a content calendar I gave her. Maybe 8 hours a week.
Lead follow-up VA: worked the CRM, sent the second and third follow-up message when a lead went quiet, updated stages, flagged anyone who'd gone cold for too long. Maybe 15 hours a week.
Add it up and that's 35 hours a week of human time, across three people, on work that is almost entirely rules plus judgement calls that don't actually need a human brain most of the time. Not all of it. Some of it genuinely did. I'll get to that.
What actually happened
I didn't wake up one day and sack three people to save money. That's not how it went and if anyone tells you it went that way for them, be a bit sceptical. What happened is slower and less dramatic. One VA left for a full-time role. Instead of hiring a replacement, I spent two weekends building the inbox and calendar workflow as an automation with an AI agent doing the triage and drafting, and me approving the sends. It worked better than the VA version within a month, because the AI never forgot a follow-up and never needed a "reminder of how I like things."
But here's the bit that actually convinced me. Not speed. Consistency. A good VA is consistent when they're switched on and thinking about your account. An AI system is consistent by default, because it's running the same process every single time, at 2am on a Tuesday same as 2pm on a Monday.
Once the first one worked, I did the same with reporting. Then lead follow-up. Each one took longer to get right than I expected, because the actual bottleneck wasn't the AI, it was me finally writing down the process properly for the first time. Turns out none of it was as documented as I thought it was. That's worth saying out loud: if you can't write the process in plain steps, no VA and no AI is going to do it well either. The documentation is the automation. The AI is just the thing that executes it without getting bored.
This tracks with what's happening more broadly. Gallup found that 47% of US employees now say their organisation has integrated AI tools into their work, up from 41% just the quarter before. That's a genuinely fast climb for something this operational. It's not a curiosity anymore, it's showing up in how people actually get their week done.
What I kept human, on purpose
And I did not automate everything, and I'd be lying to you if I said I had. Two things stayed with a human, deliberately.
Sales calls. I still do every discovery call myself. An AI can prep me brilliantly for it (and does, every morning) but the actual conversation where someone decides to trust me with their business needs a human who can read a pause on the line and change tack. I'm not handing that to a bot and I don't think I ever will.
Anything that could damage a client relationship if it went slightly wrong. Escalations, complaints, anything with money attached that isn't routine invoicing. The AI flags these to me. It doesn't handle them.
Everything else, the repeatable, rules-based, "I've done this exact thing 200 times" work, went to the system.
Where the maths actually lands
So I'm not going to pretend I tracked this to the penny, because I didn't run it as a sort of formal trial. But the honest picture: a US-based VA runs around $25.73 an hour on average as of April 2026, and even offshore VAs, where most of the market actually sits, still cost real ongoing money and management time. Stealth Agents puts the typical savings from using a VA at 60 to 78% versus a full-time US hire, which tells you the VA market itself grew up as the cheaper alternative to staff. The AI system is the next step down that same curve, not because the software is free, it isn't, but because the marginal cost of it doing one more task, or working through the night, or never needing a day off, is close to nothing once it's built properly.
The bit I underweighted going in: the time I got back wasn't just the 35 hours. It was the management overhead of running three people, checking their work, onboarding when one left, covering when one was sick. That overhead disappears entirely. Nobody tells you that part until you've lived it.
The honesty about where this still goes wrong
I'll flag the other side of this too, because I think glossing over it is how people get burned. A recent Forrester study for Boomi found that 86% of organisations have moved past pilot programmes with AI agents, but only 34% actually trust the actions those agents are taking. That gap between "we're using it" and "we trust it" is real, and it's exactly why I still have a human (me) approving anything client-facing before it goes out. The system drafts. I decide. That single rule has saved me from at least two moments where the AI got something confidently wrong.
If you're thinking about doing something similar, don't start by trying to replace a whole role in one go. Start with one task inside that role, the most repetitive one, the one that makes your VA or your own eyes glaze over. Get it running properly, watch it for a few weeks, then move to the next task. That's the only order that's worked for me. Trying to leap straight to "replace the whole job" is how people end up with a system nobody trusts, because nobody checked it properly on the way up.
If you want a look at the actual building blocks I use for this kind of thing, from prompts to workflow templates, I've put 77 of them together in the free AI Setup Kit. None of it is theoretical. It's the same stuff running my business right now.
I still think about those three VAs sometimes, not with guilt, with respect. They were good at the job as it existed. The job just doesn't exist in that shape anymore. That's not a comfortable thing to write, but pretending otherwise doesn't help anyone reading this who's wondering whether the same thing is coming for a role in their own business. It probably is. The only question is whether you build the replacement on your terms, deliberately, one task at a time, or you wait until the decision gets made for you.
Sources:
- Gallup, "Organizational AI Adoption Jumps Six Points" (2026-07-21): https://www.gallup.com/workplace/712736/organizational-adoption-jumps-six-points.aspx
- TAIO, "How Much Does a Virtual Assistant Cost in 2026" (2026-04-24): https://taio.cx/blog/virtual-assistants/how-much-does-a-virtual-assistant-cost-in-2026/
- Stealth Agents, "Virtual Assistant Statistics 2026 Full Data" (2026-05-13): https://stealthagents.com/research/virtual-assistant-statistics-2026
- Forrester Consulting for Boomi, via Insider PH (2026-07-21): https://insiderph.com/boomi-study-finds-ai-agent-trust-lags-enterprise-adoption
Want to try this yourself? Grab the free AI Setup Kit. 77 plug-and-play skills for content, sales, marketing, and operations. Works with Viktor, Claude, or ChatGPT. No course, no tool to configure, just markdown files and your AI.
About Steven Tann: Steven helps business owners build systems that run themselves using AI. After 10+ years helping 7,000+ businesses and building his own autonomous operations, he's the bloke who actually does it, not just talks about it. Find out more at steventann.com.