I'm not building an AI Agent. I'm hiring a virtual employee.

When I tried to explain to a few peers what I’m building, I couldn’t. The best I could manage was: “I’m creating virtual employees. The most cost-effective ones possible.”

That got confused looks. So let me try again — this time in writing.

The question nobody is asking

Everyone in tech is talking about AI agents. Gartner predicts 40% of enterprise apps will embed AI agents by end of 2026 (Gartner, 2025). Forrester says enterprise applications will move beyond enabling employees to “accommodating a digital workforce” (Forrester, Predictions 2026). KPMG reports 67% of business leaders will maintain AI spending even in a recession (KPMG Q4 AI Pulse Survey, Jan 2026).

But while the industry debates frameworks and ROI, I keep coming back to a simpler question:

If your entire team works remotely — communicating through Slack, reviewing PRs on GitHub, resolving tickets in Jira — what exactly separates a human colleague from an AI one?

Is it a direct message? It can do that. Group message? Sure, it can be there as well. Creating and reviewing Jira tickets? Same, both can do it. Creatin a PR, reviewin ga PR, deploying? It can do that as well. Answer questions your team gets on their Slack channel? Again… both can do it.

Think about your last Monday

You opened Slack. Messages from colleagues overnight. Someone reviewed your PR and left comments. A support ticket was answered. An alert was investigated and a brief summary posted in the monitoring channel. A Jira ticket was created for a bug someone noticed.

Now — which of those actions required a human body? Which required consciousness? Which required someone to physically exist?

None of them.

In a remote-first world, your colleagues are already abstractions. You interact with their output: messages, code reviews, tickets, decisions. The medium is text on a screen. The identity is a profile picture and a display name.

What I’m actually building

I’m building what I call virtual employees. Not chatbots. Not assistants. Not copilots.

A virtual employee has:

Its own Slack account (that’s how the team interacts with it)
Its own GitHub account (it creates, reviews and approves PRs)
Its own Gmail account
Its own Okta identity
It’s own Jira and Confluence account
Its own laptop (present in physical form)

It gets onboarded like any new hire. It gets access to the systems it needs. It operates within the same communication channels as the rest of the team.

And it works. It reviews pull requests. It investigates alerts. It answers support tickets. It creates Jira tickets and works on them. It does 1-on-1 messages with team members who have questions.

It’s not a tool you invoke. It’s a colleague you message.

The determinism spectrum

Here’s where it gets interesting for engineers.

Not everything an AI employee does needs to be non-deterministic. Most people hear “AI” and think of an LLM generating responses to every input. That’s expensive, slow, and unnecessary for a large portion of engineering work.

What I built instead is a spectrum… a hybrid of both worlds (deterministic and non-deterministic)

Every incoming request gets deterministic analysis first. The system knows which Slack channel the message came from — and channels are categorized: support, monitoring, internal. It knows whether the message starts a new thread or resumes an existing one. Based on these signals, the system determines a precise course of action: fully automated, or AI-assisted with custom prompts tailored to the context.

When deterministic analysis can’t classify the request, a lightweight AI model quickly categorizes it. If that still doesn’t resolve it, the system escalates to an actual human team member.

The result: most interactions are fast, cheap, and predictable. AI reasoning is reserved for the cases that genuinely need it.

It learns from feedback. Like any new hire.

Here’s what makes this feel less like automation and more like onboarding.

When the virtual employee reviews a PR and a human colleague disagrees — leaves a comment, requests changes, or overrides the decision — that feedback is captured. The system uses it to adjust. Next time a similar pattern appears, it handles it differently.

When the virtual employee responds to others in or team-support slack channels, we give it feedback. When it reacts on alerts, we give it feedback. And it can self-improve base don that feedback.

This is not fine-tuning a model. It’s the same loop every engineer goes through: do the work, get feedback, adapt. The difference is that a virtual employee doesn’t forget the feedback after a long weekend. And it applies it (kinda) consistently across every future interaction.

Over time, the virtual employee gets better at its job — not because someone retrained it, but because its colleagues shaped it. Just like they would with any other team member.

The cost argument

A full-time employee in Europe costs €3,500–6,000 per month in salary alone, before benefits, equipment, and office overhead (Hoeijmakers.net, 2025). An AI agent performing comparable task-based functions costs €100–300 per month in compute, hosting, and API costs (Hoeijmakers.net, 2025).

But cost isn’t even the main point. The real shift is in scalability. You can’t clone your best on-call engineer. You can clone a virtual employee that handles 80% of alert triage at 3 AM without anyone losing sleep.

Why most “AI agent” projects will fail

Gartner also predicts that over 40% of agentic AI projects will be cancelled by the end of 2027 — due to escalating costs, unclear business value, and inadequate risk controls (Gartner, June 2025). They estimate only about 130 of the thousands of “agentic AI” vendors actually offer real capabilities. The rest is what Gartner calls “agent washing” — rebranding chatbots and RPA tools.

I think the reason is simpler than vendor hype. Most AI agent projects fail because they treat AI as a tool, not as a team member. They bolt a chatbot onto existing workflows and expect transformation. Imagine hiring a consultant and expecting a full organization transformation just because you’re paying for them. The only that will change is the number in your “costs” column.

A virtual employee is different. It’s designed from the ground up to operate like a person: with identity, access, context, and accountability. It doesn’t wait to be invoked — it participates.

The uncomfortable question

44% of business leaders now expect AI agents to take lead roles in managing projects alongside human teams within 2-3 years (KPMG Q4 AI Pulse Survey, Jan 2026). Forrester predicts the top five HCM platforms will offer “digital employee management” capabilities in 2026 (Forrester, Predictions 2026).

We’re heading toward a world where HR systems track both human and digital employees. Where “headcount” includes agents. Where onboarding a new team member might mean provisioning an AI, not interviewing a person.

This isn’t science fiction. Salesforce reported that 83% of customer service queries on their Agentforce platform now resolve entirely without human intervention (Salesforce Q3 FY26 Earnings, Dec 2025).

The question is no longer “can AI do the work?” It’s: “if it does the work, communicates like a colleague, and operates within the same systems — what makes it not an employee?”

My answer, for now

The difference between a virtual employee and a human one is shrinking fast. But it hasn’t disappeared.

A virtual employee doesn’t have intuition born from years of domain experience. It doesn’t build relationships over lunch. It doesn’t mentor junior engineers through career crises. It doesn’t push back in a meeting because something feels wrong even when the data says otherwise.

What it does is handle the 80% of work that is structured, repetitive, and well-defined — so that humans can focus on the 20% that actually requires being human.

And yes, sometimes AI will make mistakes, you will need to steer it and spend time improving it. You will have to teach it how to behave in your company, providing it knowledge, expectations and feedback. Doesn’t that sound familiar, almost as if… ah yes… it’s what we do today with human employees.

Sources

Gartner. (June 2025). “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027.” gartner.com
Gartner via Motley Fool. (Feb 2026). “40% of Enterprise Apps Will Embed AI Agents by End of 2026.” fool.com
Forrester. (Nov 2025). “Predictions 2026: AI Agents, Changing Business Models, And Workplace Culture Impact Enterprise Software.” forrester.com
KPMG. (Jan 2026). “AI at Scale: Q4 AI Pulse Survey.” kpmg.com
Hoeijmakers.net. (June 2025). “What Does an AI Agent Cost Compared to a Human?” hoeijmakers.net
AI World Today. (Dec 2025). “Why 2026 Is the Year of the Digital Employee.” aiworldtoday.net
Kore.ai. (Feb 2026). “AI Agents in 2026: From Hype to Enterprise Reality.” kore.ai