Claude Code Gets Fallback Models and Safer Defaults
We break down Anthropic’s latest Claude Code updates, including ordered fallback models for handling rate limits, new controls for disabling thinking tokens, and stronger permission prompts around sensitive system files.
Also covered: the handy /btw clipboard command and why these small workflow improvements make agentic terminal work smoother and safer.
Is this your podcast and want to remove this banner? Click here.
Chapter 1
Claude Code Resilience with Fallback Models and Custom Configuration
Lachlan Reed
Welcome to the show, everyone! I'm Lachlan Reed, coming to you from my backyard shed, and I'm joined by James Turner. We've got a massive tech hook for you today, and thanks to Jellypod to help make this daily show a reality. If you've been running Claude Code for agentic workflows, you know the absolute gut punch of hitting API rate limits right in the middle of a complex deploy. Well, Anthropic just threw us a massive lifeline. [excited]
James Turner
Oh, they absolutely did. [excited] We are talking about true multi-model resilience now. You can now configure up to three fallback models in strict order using the new `fallbackModel` setting in your config file, or just append the `--fallback-model` flag on the fly. If Claude 3.5 Sonnet starts throwing 429 rate-limit errors or peak-traffic overloads, your agentic terminal loop will immediately fall back to your secondary or tertiary choice without crashing the run.
Lachlan Reed
That is fair dinkum brilliant. [chuckles] No more babysitting a script only for it to fall over at 2:00 AM because of API congestion. You could literally set Sonnet as primary, fallback to Haiku for speed, and then maybe keep an older model as a safety net. It's beautiful. But wait, James, if we're falling back to models that have deep reasoning capabilities, isn't that going to run up the bill with thinking tokens? [questioning tone]
James Turner
It definitely can, which brings us to the second major configuration change: thinking toggles. For workflows where you do not need deep chain-of-thought processing, you can completely disable thinking by setting `MAX_THINKING_TOKENS=0` in your environment, or simply run `--thinking disabled`. [matter-of-fact]
Lachlan Reed
Disable it entirely? [thoughtfully] That makes a lot of sense for simple, structured tasks. Like, if I'm just regex-parsing some logs, I don't need Claude spending five seconds and thousands of tokens pondering the meaning of life before it writes the script. It's a huge latency and cost saver.
James Turner
Exactly. [matter-of-fact] Save the heavy thinking for complex logic, and bypass it for routine file manipulations. And speaking of file manipulations, the latest security prompt updates are a massive deal. Claude Code now has explicit guardrails around critical system files. It prompts for explicit user permission before touching things like your SSH keys, shell profiles, or global env files.
Lachlan Reed
No worries about an agent accidentally nuking your `.bashrc` or sending your private keys off to a remote server then. [laughs] That's a massive relief for local execution. And while we're on local workflows, have you seen the new `/btw` command? It copies the last command output directly to your system clipboard. [curious]
James Turner
Oh, the clipboard copy! [excited] It's such a simple quality-of-life addition, but not having to drag-select text in a messy terminal window is a game-changer. You just type `/btw` and paste it right into your documentation or Slack.
Lachlan Reed
It's the little things that keep us sane, mate. Keep those fallbacks configured and your system files locked down. That's all for today's quick tech breakdown. Catch you next time! [warmly]
James Turner
See ya! [cheerful]