5 Simple Things to Get Your Business Ready for AI Agents

I was on a live implementation call with the president and CEO of a company, putting AI agents into his business for the first time. Almost everything that could block us did, and none of it was the AI. It was the five things sitting underneath the AI, and they have to happen in order. Here they are, in the sequence they actually unfold.

1. Be open to a grueling first eight to ten hours

Before any of the technical work, there is a decision: are you willing to sit through the boring part. Across the 10-plus businesses I have helped implement agents, the ones where the CEO goes first move the fastest. It is the only way I will run an implementation, because a CEO does not feel what is missing until they put in the first six to eight hours themselves and hit the walls in person.

Go in knowing the first eight to ten hours are going to be brutal. Boring, hard, monotonous, frustrating. The reward is real, but it comes after the grind, not during it. Get through that first stretch and you move faster than anyone, far enough ahead of your competitors that it stops being funny, because everything you set up next can be multiplied across every team member. Nobody on your team can do this part for you. The whole environment was built for humans, and only the person sitting in it feels where it binds.

2. Replace the software your agents cannot operate

Once you are in the trenches, you start to see it: the vendors and software running your business were chosen for people, not agents. This is the step you cannot appreciate from the outside. You only feel its weight when an agent tries to do real work and the tools you pay for will not let it.

Here is how it showed up on the call. The first thing we tried was installing Claude Code, and his own IT company blocked it, with no admin override available. This is the firm he pays. He called them from the Zoom, refused to be routed to a support ticket, and pulled rank as the CEO just to approve a basic install. An IT setup that flags everything coding-related for super-admin approval was built for human users clicking known apps, not for an agent that needs to act. A vendor that cannot support agents, and has no agent-readiness offer to sell you, is holding your business back. He is firing them and finding one that can grant the safe, scoped permissions agents need.

The same logic runs through your whole stack. Any closed, all-in-one platform an agent cannot reach has a ceiling, no matter how convenient the single login feels. Be willing to move onto future-forward software an agent can work inside, the kind of open infrastructure where the work is reachable. Until your tools can let an agent in, nothing downstream matters.

3. Invest in the hardware

Within the first few prompts on the call, his computer started to crawl. The reason was 8GB of RAM, which is not what you want on a machine meant to run Codex or Claude Code, and it gets far worse once several agents are working at once. Treat the machine as an investment in the business, not an afterthought.

You do not have to guess where yours stands. Paste this prompt into Codex or Claude Code and it will audit the hardware of the computer you are working on right now. It is read-only, so it only inspects and reports, and it works on Mac, Windows, or Linux.

You are an expert at auditing whether a computer can handle running AI coding agents. I need you to inspect this machine and tell me, in plain English, whether it can comfortably run terminal coding agents like OpenAI Codex or Claude Code, including running several of them at the same time.

STRICT SAFETY RULES, follow these exactly:
- This is READ-ONLY. ONLY inspect and report. Do NOT install, download, update, change, move, rename, or delete anything. Do NOT modify any settings or files. Do NOT run package managers or any command that writes to disk. If a check would require any of that, skip it and say so.
- Only run standard, built-in commands that read system information. If you are unsure whether a command is safe, do not run it.

STEP 1: Detect the operating system first (macOS, Windows, or Linux), then use the correct commands for that system. Examples you may use:
- macOS: `sysctl hw.memsize` and `vm_stat` for RAM, `sysctl -n machdep.cpu.brand_string` and `sysctl -n hw.ncpu` for CPU, `df -h /` for disk, `sw_vers` for OS version, `ps` sorted by memory for top processes.
- Windows (PowerShell): `Get-CIMInstance Win32_OperatingSystem` for total/free RAM and OS version, `Get-CIMInstance Win32_Processor` for CPU model and cores, `Get-PSDrive C` or `Get-Volume` for disk, `Get-Process | Sort-Object WorkingSet -Descending` for top processes.
- Linux: `free -h` for RAM, `lscpu` or `/proc/cpuinfo` for CPU, `df -h /` for disk, `cat /etc/os-release` for OS version, `ps` sorted by memory for top processes.

STEP 2: Inspect and collect all of the following:
- Total RAM and currently available/free RAM
- CPU model and number of cores
- Total disk space and free disk space on the main drive
- Operating system name and version
- The handful of heaviest running processes by memory use

STEP 3: Judge readiness for running AI coding agents using these rules:
- 8GB RAM total: agents will crawl, especially with a browser open. Not recommended for real work.
- 16GB RAM total: workable floor for one agent, maybe two if you keep other apps closed.
- 32GB RAM or more: comfortable, including running several agents at once.
- Free disk space matters: less than about 10GB free is a problem, 50GB or more free is healthy for code, dependencies, and models.
- Also weigh how much RAM is actually free right now and whether heavy apps are already eating memory.

Give a single clear verdict from these three options: Ready, Marginal, or Upgrade needed.

STEP 4: Report back in this exact shape and nothing more:
1. A short table with these rows: Operating system, Total RAM, RAM free now, CPU, CPU cores, Disk total, Disk free, Heaviest apps now.
2. One line: "Verdict: Ready" or "Verdict: Marginal" or "Verdict: Upgrade needed" for running one agent, and a note on whether running several at once is realistic.
3. One line naming the single most important upgrade or change, if any (for example "Add more RAM, aim for 32GB" or "Free up disk space" or "Nothing needed, you are good").

Keep the language plain and friendly. No technical jargon the average business owner would not recognize. Do not suggest any commands for me to run. Do everything yourself, read-only, and just give me the report.

Run it, then buy the machine the work needs. 16GB of RAM is a workable floor, 32GB is comfortable for multiple agents, and the cost of the slowdown compounds until you wish you had upgraded at the start. While you are here, graduate off plain ChatGPT chat and commit to one agent. Most of my clients start on Codex. Chat answers questions; it does not act, and acting is the entire point.

4. Connect Codex and Claude Code to the right environments

Only now, with agent-ready software and a machine that can keep up, does it make sense to connect your desktop agents to the systems where the work lives. You do this through plugins: the agent links to your email, your calendar, your code repository, your CRM, and any other core tool it needs to act inside.

On the call, even this got blocked. When we went to connect Microsoft Outlook email and calendar, GitHub, and HubSpot, the browser login path that authorizes each plugin was being blocked by his IT firm. You install a plugin, a browser is supposed to open, you log in, and that login is what grants the agent scoped access to that environment. Block the login and the agent has no access, which means it cannot do the core tasks you hired it for. This is exactly why the software step comes before this one. If you are new to this, start with what a plugin really is, then install your first Codex plugin.

5. Give them permission and instructions to run the work

The last step is where the agent stops talking and starts operating. You grant it permission to act on its own instead of stopping to ask at every step, and you hand it clear instructions for the jobs you want done. On the call, we set the agent up so it could truly act agentically, then gave it the prompts I had prepared so it could start working through real tasks in his business.

This is where the reward from step one finally shows up. Every agent reports its own actions to the millisecond: which project it touched, the progress made, the items still pending. His agents in Codex and Claude Code do it, and so does every other employee's, which gives the CEO real-time insight into the business as it runs. He wants every employee managing at least one agent within six months. Once a leader has seen that, the policy is easy to set: standardize on Codex and Claude Code, and stop scattering the team across private chat windows you can never learn from. From there the next level is proactive analysis, the agents surfacing what matters before anyone asks.

Quick Answers

What are the steps to get my business ready for AI agents? Five, in order. Be open to a grueling first eight to ten hours. Replace the vendors and software your agents cannot operate inside. Invest in hardware that can keep up. Connect Codex and Claude Code to your real environments through plugins. Then give the agents permission and instructions to start running the work.

Why do I have to replace my current software to use AI agents? An agent can only operate inside tools that give it access. A closed vendor or all-in-one platform an agent cannot reach has a ceiling no matter how convenient it is. You usually do not see which tools are blocking you until you are in the trenches doing the work.

How much RAM do I need to run Codex or Claude Code? 8GB will crawl, especially with a browser open. 16GB is a workable floor for one agent, and 32GB or more is comfortable for running several agents at once.

How do AI agents connect to my email, calendar, and other tools? Through plugins that authorize scoped access. You install the plugin, a browser opens, and you log in to grant the agent access to that system. If your IT setup blocks that login path, the agent cannot connect, so the infrastructure has to be fixed first.

Should I use ChatGPT, Codex, or Claude Code for AI agents? Plain ChatGPT chat answers questions but cannot act inside your systems. For real agent work, commit to a terminal coding agent like Codex or Claude Code. Most owners are best starting on Codex.

Why should the CEO go first instead of delegating? Only the person sitting in the work feels where the business binds. Across more than ten implementations, the rollouts where the CEO goes first move the fastest, because the CEO sees what has to change and can set policy on the spot.

The real question to ask

So here is the question I left him with. If every one of your top clients and prospects is running at least one capable agent in Codex or Claude Code within 18 to 24 months, which of your services loses value, and what are you going to do about it?

Want to see this built on a real business, step by step? Join the next free AI masterclass, or have us run the implementation with you at Growth Academy.