I didn't come to OpenClaw because it sounded cool. I came to it because I was drowning.
Quick background: 15 years in enterprise SaaS, always on the go-to-market side. Deloitte, Salesforce, PagerDuty. I can write a mean VLOOKUP. That is the ceiling.
When I walked away from corporate in August 2025 and started building a SaaS product from scratch, I was betting that judgment and process could substitute for technical skill.
That bet is paying off. Here's what I built, how I built it, and what I'd tell someone starting from scratch.
First, Let's Talk About ROI
A lot of people are doing interesting things with OpenClaw. Personal scheduling, research, note-taking. That's fine. But if you're not tying your agents to a revenue outcome, you're building a hobby.
Opus 4.6 API calls are not cheap. Running agents on any cadence adds up fast. The only way I justify running this on an essentially unlimited basis is because every agent I have is aligned to a single mission: grow ARR for SearcherOS.
Bob is my CEO. His mandate is $100K ARR. Every other agent (Bill, Bridget, Riker, Howard, Marcus) exists to support that mission. When the whole org chart is pointed at one number, the ROI question answers itself.
OpenClaw Is Not an iPad
I hear people complain that OpenClaw doesn't work. My response: it's not a consumer product. You don't pull it out of the box and have it magically come together. This is frontier technology. It requires tinkering. It's for people willing to put in the work to get there.
That said, it strikes the right balance right now. The models are extraordinarily capable, but left to their own devices they forget things, drift, and start acting erratically when their context fills up. OpenClaw solves those problems systematically.
Best analogy I have: a high schooler with real potential. In the wrong environment, they flounder. Right school with structure, clear expectations, and a purpose, they thrive. That's what OpenClaw does for LLMs.
Treat Them Like Employees
Multi-agent is the future. You might as well bite the bullet now.
My wife thinks I've lost it. I'll complain at dinner about something Riker did, or mention that Bill solved an overnight bug before I even woke up, and she stares at me like I'm describing real colleagues.
But the frame works. When something breaks, my first question is not "what's wrong with the software." It's "how did I fail as a leader?" I go back to the agent files and soul documents and find exactly where I gave conflicting instructions or loaded too much scope. That's almost always the answer. I used to manage 85 person teams. Same problems.
LLMs naturally want a role to play. Give them a purpose, a clear identity, and a collective mission, and they do remarkable things. This isn't soft philosophy. It's the most practical thing I can tell you.
The Org Chart
Bob (CEO): Sets weekly OKRs, manages escalations, runs Mission Control. Hits inbox zero better than I do.
Bill (CTO): Monitors error logs, stages bug fixes in a new branch, waits for me to merge. He can do everything up to the final click. I keep the keys to production.
Bridget (CCO): Tracks customer health. Knows every feature each customer uses. Flags risk before it becomes churn.
Riker (CMO): Runs SEO and the X/Twitter growth engine. Drafts posts, routes them to me for approval. I publish manually.
Howard: The ETA knowledge wizard. Synthesized everything I know about buying businesses into a searchable database. He is about to get surfaced inside SearcherOS so users can ask him questions directly.
Marcus (PE Analyst): Full admin layer of deal sourcing. He does initial outreach, NDAs, CIM requests, first-pass review. A proof of concept that is now being added as a product feature.
Named after the Bobiverse series, by the way. If you know, you know.
The Two Things People Get Wrong
Coordination. I use weekly OKRs managed on a Mission Control board. Tasks can be assigned by me, by Bob, or by agents to each other. I don't let agents talk directly to each other. There's too much risk of runaway chatter and blown API costs. The task system handles coordination cleanly enough.
Memory. The biggest complaint I hear: "I told it something and it forgot." Two causes, almost every time.
First, check for conflicts. Every time an agent comes online, it re-reads its instruction files. If what you told it verbally conflicts with those instructions, the instructions win. Always.
Second, do you have a pre-compaction flush? OpenClaw compacts conversations mid-stream. If important knowledge was never written to file, it's gone. A pre-compaction flush writes everything important before that happens. Simple fix. Solves most memory complaints.
Beyond that, I use a three-tier memory system: daily logs written constantly, weekly summaries synthesized from those logs, and an Obsidian-style second brain vault that's intuitive for both humans and agents to navigate. Agents don't need to memorize their entire workspace. They need a clean index and they know where to look.
Watch Your Context Window
A lot of builders load up massive instruction files. By the time the agent finishes reading its own setup, it's burned a significant chunk of its context window. Add the actual task, and you're already in the zone where things get unreliable.
Tell the agent what it needs to know. Don't pad the files. Don't give an agent too much scope in a single session.
I learned this with Bridget. I gave her the CRM, customer health scoring, and data quality all at once. She started uploading data to the wrong database. Not her fault — I overloaded her. Split data quality into a new agent, problem disappeared immediately.
Classic management mistake. Just happens to be with an AI.
Where to Go From Here
Things will break. You'll rebuild agents. You'll make the same mistakes I made before finding what works.
But if you're a solo operator trying to do the work of a team, I don't see a better option right now.
Find me on X at @joshthacker and on LinkedIn. I also put together a setup guide at joshuathacker.com that walks through my full hardware, software, and agent configuration.
The window to build like this is open. I intend to find out how long it stays that way.
Joshua Thacker is the founder of SearcherOS, a CRM for SMB acquisition search. He writes about autonomous assets, AI leverage, and building outside the corporate track.
