I've been moving away from OpenClaw for weeks. Auth tokens expiring, gateway errors mid-task, scheduled jobs that looked healthy until they weren't. I wrote about this in my last two posts.
Then Anthropic sent an email saying Claude subscriptions no longer cover OpenClaw usage. Starting today.
So that decision I'd been putting off? Made for me.
The problems were stacking up before the email
This wasn't a sudden breakup. OpenClaw had been giving me friction for months. The issues were always the same boring category:
- Auth tokens expiring without warning
- Cron environments losing variables between runs
- Gateway errors killing jobs mid-execution
- Scheduled automations that needed constant babysitting
- Jobs that reported healthy status but produced nothing
I was spending more time diagnosing failures than getting work done. That's backwards. The whole point of an agent is to remove work from your plate, not add a new category of maintenance.
For experiments and quick one-off tasks, OpenClaw was fine. For anything I needed to run reliably on a schedule, it kept letting me down.
What Anthropic actually said
Boris Cherny from Anthropic posted the details:
The short version: Claude subscriptions (including Max) will no longer cover third-party harness usage. OpenClaw, Lindy, anything that isn't Claude Code or Claude Cowork gets cut off from your subscription.
You can still use OpenClaw with an API key or with extra usage bundles (pay-as-you-go, billed separately). They're offering a one-time credit equal to your monthly subscription and discounts up to 30% on prepaid bundles. Refunds available if you want out.
The reasoning is straightforward: third-party tools put outsized strain on their systems. They want to prioritize capacity for their own products.
The Codex/ChatGPT fallback is not the same
The first reaction from most people was "Alright, switch to Codex." I tried that. Switched my OpenClaw to ChatGPT's models on the subscription.
It was noticeably worse. The quality gap between Opus/Sonnet and GPT-5.3-Codex for the kind of agentic work I was doing (multi-step automations, code generation, context-heavy reasoning) was real. Tasks that Claude handled cleanly required more hand-holding on Codex. More corrections, more re-runs, more supervision.
Some people in the community report good results with Codex for simpler tasks. For code completion and straightforward scripting, it's solid. But for the kind of autonomous agent work where you describe a problem and the model builds a full solution, Claude is still ahead. That's not a knock on OpenAI. It's just where the models are right now.
So the "just switch to Codex" advice doesn't fully hold if you're running real workloads.
What it actually costs on the API
Here's where it gets concrete. I pulled my actual Anthropic billing data from February when I was running OpenClaw heavily on the API.
My daily API spend looked like this:
| Date | Cost | |------|------| | Feb 13 | $48.06 | | Feb 14 | $48.26 | | Feb 15 | $48.18 | | Feb 16 | $48.36 | | Feb 17 | $48.07 | | Feb 18 | $49.51 | | Feb 19 | $48.14 | | Feb 20 | $48.19 | | Feb 21 | $49.09 | | Feb 23 | $48.51 |
That's roughly $48/day. For a full month of heavy OpenClaw usage on the Anthropic API, you're looking at around $1,400-$1,500/month.
Here's the cost comparison depending on how you use it:
| Usage level | Claude Sub (before) | API (now required) | Codex Sub | |-------------|--------------------|--------------------|-----------| | Light (1-2 hrs/day) | $20/mo (included) | ~$200-400/mo | $20/mo | | Moderate (4-5 hrs/day) | $20/mo (included) | ~$600-900/mo | $20/mo | | Heavy (8+ hrs/day) | $20/mo (included) | ~$1,200-1,500/mo | $200/mo | | Power user (always on) | $20/mo (included) | ~$1,500-2,000+/mo | $200/mo |
The price jump from "included in your $20 sub" to "$1,500/mo on the API" is why people are panicking. For heavy users, this is a 75x price increase overnight. Even with the 30% discount on prepaid bundles, you're still looking at $1,000+/mo.
If you're a power user, you're either eating that cost, downgrading to Codex (with the quality tradeoff), or moving to Claude Code where your subscription still works.
What I moved to Claude Code
Over the past month I migrated my critical automations one by one. The biggest was my Granola meeting notes backup system, which I wrote about in detail. But I also moved:
- Daily Google Ads reporting (the n8n workflow orchestration)
- GitHub repo sync and backup automation
- Email monitoring and daily agent briefs
- Scheduled heartbeat checks
All of these now run through Claude Code with native macOS scheduling (launchd). No npm dependency mess. No orchestration layer that can break independently. Just scripts, cron, and GitHub as permanent memory.
The reliability difference was immediate. Jobs that failed weekly on OpenClaw have been running without intervention on Claude Code for a month.
What OpenClaw was actually better at
I want to be honest about this because it wasn't all bad.
Channel integrations. Having an agent live in Slack or Discord meant async collaboration across a team. You could message it, it would respond with context from previous conversations. That interaction model is more natural than opening a terminal.
Persistent sessions. If you're chaining 15+ autonomous steps, OpenClaw handled that without dropping context as aggressively. Codex drops context even more, which makes it worse for this use case.
Mobile access. You could manage agents from your phone through the chat interface. That's something Claude Code doesn't match out of the box.
Fixing the mobile gap: Claude Code on WhatsApp
This was my biggest frustration. I use the Claude iOS app constantly, but Claude Code doesn't pull in the conversations I'm having there. They're separate contexts. When I'm on my phone and want to check on a scheduled job or trigger a workflow, I can't. I have to wait until I'm back at my laptop.
On top of that, you have to manually click into Claude Code within the app rather than it being the default mode. If you're already running everything through Claude Code on desktop, the mobile app should feel like the same workspace.
So I built a bridge. A simple WhatsApp bot that connects to Claude via the Anthropic API. It takes about 10 minutes to set up with Twilio and a basic Node.js server.
The setup:
- Twilio account with WhatsApp sandbox (free to start)
- A small Express server that receives WhatsApp messages via webhook
- Forwards them to Claude via the Anthropic SDK
- Sends Claude's response back to WhatsApp
It maintains conversation history per phone number, so context carries across messages. It's not full Claude Code with file access and tool use, but for quick questions, status checks, and kicking off tasks, it works.
I open WhatsApp, message my agent, get a response in seconds. Same as I used to do with OpenClaw's chat interface, but running on my own infrastructure with my own API key.
The code is in my nickbuilds-whatsapp repo if you want to set it up yourself.
What I still want from Anthropic: bring Claude Code context to the mobile app natively. Same session, same tools, same memory. The WhatsApp bridge is a workaround. A real mobile-native Claude Code experience would close the gap completely.
What I'd tell someone starting fresh
Skip OpenClaw. Start with Claude Code.
Build your agent's memory in GitHub. Use skills files to define what it knows and how it works. Use native scheduling (launchd on Mac, cron on Linux) for recurring tasks. If you need mobile access, set up the WhatsApp bridge or wait for Anthropic to ship mobile Claude Code.
The agent ecosystem is shifting fast. Anthropic is making it clear they want you on their own tools. Building on someone else's relationship with the model provider was always a risk. Now that risk materialized.
Own your stack. Keep it boring. Make it reliable.