By Eru Penkman, Identity Software
I’m Eru Penkman. I run Identity Software, a small custom software agency on the Sunshine Coast and in Brisbane. We build business apps that replace spreadsheets, paper and clunky off-the-shelf tools. Like most small agencies, our marketing usually gets whatever time is left over. So I’ve just set up two AI agents to take it on: Grok Bot and Claude Code. I expected to pick a winner. Instead, I gave each one a different job.
Grok Bot: the workforce
Grok Bot feels like a team member, and it was quick to set up. You chat with it in plain language and it gets on with the job, with no terminal and no code reviews.
It’s best at ongoing work that never really finishes: checking my Google Ads a few times a week, blocking wasted spend, testing analytics, installing WordPress plugins and making small site updates. Because it runs on a schedule and remembers what it did last time, it keeps improving things without me having to ask.
Connecting it to other services was also much easier. Grok Bot plugs into Composio, an integration platform, so linking Google Ads and Google Analytics took a few clicks. With Claude Code, I tended to wire up each external service by hand.
It isn’t reckless either. It can pause things and cut spend on its own, but anything bigger comes to me for approval first.
My recommendation: for day-to-day work, with a team you can simply talk to and get going quickly, start with Grok Bot.
Claude Code: the infrastructure and security engineer
Claude Code has deeper access and stronger technical skills. For example, it can connect to my server over SSH.
It’s also far more conservative. It insisted on a full pull request for every small change to the live site, which is good practice for software but slows marketing right down. At times it felt like working with a brilliant but grumpy engineer.
That caution is exactly what you want for the jobs that need real care: initial setup, security hardening, firewall changes, server updates, a website migration or a complete redevelopment. Those are mostly one-off projects you’d want to supervise closely anyway.
My recommendation: for setup, infrastructure, security-critical work and one-off technical jobs, use Claude Code.
How I’ve split the work
The way I picture it, Grok Bot is the workforce and Claude Code is the infrastructure and security engineer.
On the Grok Bot side, I’ve set up two agents that each have a job, like team members:
- A marketing manager that runs our Google Ads, watches the search terms and spend, and reports back to me.
- A website manager that works like a WordPress developer, handling content, plugins and site updates.
They each have their own chat, and they talk to each other when a job crosses over.
Claude Code sits behind them. Like a good infrastructure engineer, it’s essential, but I don’t deal with it much day to day. When the team needs a server change, they tell me exactly what’s needed and I pass it on.
There isn’t yet a good, popular way for the Grok Bot agents to talk to Claude Code directly. In practice that hasn’t mattered much. The Grok Bot team makes most decisions on its own, and the rare server-level change is exactly the kind of thing I want a human to look at first.
The first real test went well. The marketing manager suggested adding session analytics, and the website manager set it up, drafted the privacy policy update for my approval and deployed the changes. Then they handed me two server steps to give to Claude Code.
Where this is heading
Claude Code can now run scheduled work in the cloud too, through cloud scheduled tasks (they used to be called routines). Those jobs run in a fresh cloud virtual machine that can edit a GitHub repo and push a branch, so you don’t need your laptop on, but that environment isn’t a lasting workstation: each run starts from a fresh clone, and it doesn’t keep a desktop or a signed-in browser between jobs. With Grok Bot, my agents share one cloud computer that holds files, tools and logins, and each agent has its own desktop on it, which is how I can change the website entirely in the cloud.
I think this is where things are heading, and longer term I’d like each agent in its own isolated environment. My website agent and an agent coding on client projects shouldn’t be able to see each other’s work. That matters even more with Claude Code on my laptop, which can reach almost anything on my computer and is held back mainly by permission rules rather than a hard boundary.
At a glance
| Grok Bot | Claude Code | |
|---|---|---|
| Think of it as | The workforce | The infrastructure and security engineer |
| Best for | Day-to-day marketing and website work | Setup, infrastructure, security-critical and one-off tasks |
| How you work with it | Chat, like a team member | Terminal and code reviews |
| Ongoing tasks | Runs on a schedule and remembers past work | Can run on a schedule in the cloud, but each run starts fresh |
| Connecting other services | A few clicks through Composio | Mostly set up by hand |
| Server access | No SSH | SSH and full server admin |
| Approach to change | Acts within the rules you set | Very cautious, pull request for every change |
The takeaway
It’s not about which AI is best. It’s about matching each tool’s strengths and safeguards to the job, with a human coordinating the hand-offs.
For the day-to-day, start with Grok Bot. For setup, infrastructure and anything security-critical, bring in Claude Code.
This is a snapshot in time. These tools are changing very quickly, and I expect this split to look different in a few months.
Thinking about how AI or custom software could take work off your plate? Book a free 45-minute Vision Session and we’ll talk it through.