Agentic AI: When You Actually Need It, and Platforms You Can Run Directly on a VPS

Over the past year "agentic AI" has become one of the most overused terms out there, sometimes for something genuinely agentic, sometimes just for a chatbot or plain automation rebranded. On one hand, adoption really is growing fast. On the other, plenty of projects end up canceled because they were built for the wrong problem.

In this article I cover two things: when you actually need agentic AI, and when it's better to hold off, plus self-hosted platforms you can now plug straight into your own VPS, including two that are getting a lot of attention right now: OpenClaw and Hermes Agent.

Agentic AI Is Different From a Chatbot or Plain Automation

Plain automation (a script, Apps Script, an n8n workflow) runs a path decided in advance: if A happens, run B. Agentic AI is different because it decides the next step itself: which tool to use, where to look for information, whether to retry a failed step, or when to ask for confirmation. It usually also carries memory across sessions, so it doesn't start from zero every time you talk to it.

The difference shows clearly in how each one behaves: plain automation stops the moment something unexpected happens. Agentic AI tries another path first before giving up.

Why It's Such a Big Deal Right Now

This isn't just a feeling. According to Gartner, 40% of enterprise apps are projected to feature a task-specific AI agent by the end of 2026, a sharp jump from under 5% in 2025.

The hype shows up at the personal level too, not just enterprise. Two open source projects you can run directly on your own VPS, OpenClaw and Hermes Agent, have both crossed hundreds of thousands of GitHub stars within months of launch.

When You Actually Need Agentic AI

A few signals that tell me it's time for agentic AI, not just plain automation:

• The task needs many steps with conditional branching, not a fixed linear flow.
• You want to be able to ask for the work from a channel you already use daily, WhatsApp, Telegram, or Slack, instead of opening a new dashboard every time.
• Context and memory across sessions matter, for example the agent needs to remember past conversations or preferences.
• The volume is large and regular enough that it's worth automating with something that can make its own decisions, not just run manually now and then.

When It's Better to Hold Off

This is the part vendors rarely talk about. According to Gartner, more than 40% of agentic AI projects are expected to be canceled before the end of 2027, due to spiraling costs, unclear business value, or inadequate risk controls. Gartner even said that out of the thousands of vendors claiming to sell "agentic AI", only about 130 genuinely have agentic capability by their assessment. The rest is just chatbots or RPA rebranded, a pattern they call "agent washing".

Projected enterprise apps with a task-specific AI agent by end of 2026, up from <5% in 2025

40%

Projected agentic AI projects to be canceled before end of 2027

>40%

Vendors Gartner judged genuinely 'agentic', out of the thousands claiming it

~130

If your workflow is actually fixed and doesn't need much conditional decision-making, plain automation is cheaper and easier to maintain. I go into that trade-off in more depth in Not Everything Needs AI.

Platforms You Can Run Directly on Your Own VPS

Once you're confident you need agentic AI and want full control over the data, not just someone else's cloud service, these two open source platforms can be installed directly on your own VPS.

OpenClaw

OpenClaw is an open source personal AI assistant, MIT licensed, maintained by the nonprofit OpenClaw Foundation. It connects to more than 20 chat channels people already use: WhatsApp, Telegram, Slack, Discord, iMessage, Signal, Microsoft Teams, and a few others, through a single gateway that runs on your own infrastructure. The model can be Claude, GPT, or a local model, depending on the provider you pick. Because OpenClaw has direct access to the terminal and file system, its own documentation recommends running it on a dedicated device or VPS that doesn't hold other sensitive data.

Hermes Agent

Hermes Agent was built by Nous Research, released on February 25, 2026, also open source under an MIT license. Unlike OpenClaw, Hermes is designed to be "self-improving": it creates and stores new skills from experience, has memory it can search across sessions, and comes with a built-in cron scheduler for timed tasks. It can run on a VPS, a GPU cluster, or serverless, and connects to Telegram, Discord, Slack, WhatsApp, Signal, and even email. Its model support is flexible: Nous Portal, OpenRouter, OpenAI, Anthropic, or a local model through Ollama if you want costs close to zero. Since July 2026 there's also an official hosted option, Hermes Cloud, if you'd rather not manage the VPS yourself.

GitHub popularity(thousand stars)

• OpenClaw~390K
• Hermes Agent~246K

Both are still young, fast-moving projects, so treat this as a starting point for your own research, not a final recommendation. Check the official docs for the latest specs and install steps before you start.

What to Set Up Before You Start

• Reasonable VPS specs. I go into specs and pricing in more depth in What Is a VPS for AI, and What Can It Actually Do?, but the rule of thumb is similar: most automation is fine on 2 vCPU and 4 GB RAM, browser automation needs more.
• Keep it separate from production data. Don't put an agent with full terminal and file system access on a server that also holds other sensitive data.
• Start with one narrow task. Expand only once it's actually working well, instead of handing a freshly installed agent a lot of decisions at once.

Wrapping Up

Agentic AI is worth trying if the task genuinely needs a lot of conditional decision-making and you want it to run from a channel you already use every day. But don't start from "I want to use agentic AI", start from the real problem you're trying to solve, then decide whether it needs an agent that can make its own decisions, or whether plain automation, which is far cheaper to maintain, is enough.

If you're still weighing which one fits your needs, tell me about your automation or agentic AI needs, and we'll figure out the right approach together.