The short answer
Pick Claude Code if you want speed and a tight feedback loop while you are still figuring out what to build. Pick OpenAI Codex on extra-high reasoning when you want fewer bugs and production-grade polish, and can wait five to ten minutes per run.
That is where we landed after switching back and forth at least three times in a year. Jordan's daily driver is Codex. Sam splits roughly 50/50, and his recent builds start in Claude Code and finish in Codex.
What we built with each
Most of what we know comes from shipping real tools on the show, not benchmarks.
Claude Code builds
- Sam's ATS: Sam's first time using Claude Code. He built a full applicant tracking system front end in about an hour, running up to three agents in parallel (Episode 9).
- Kalshi prediction market trading bot: Jordan's full-stack bot, built with Opus 4.5, Cursor and Claude Code over the holiday break.
- ScreenEval: a recruiter coaching app Sam built in about six hours in the Claude Code terminal with Supabase, without ever having written a line of code.
- Personal DNA and health insights dashboard: Jordan gave Claude Code his raw 23andMe file and blood work; it runs locally on SQLite.
Codex builds
- Strudel AI DJ mixer: Jordan said Codex built "the majority" of the two-deck mixer (Episode 11, at 14:45).
- Rep-by-rep AI personal trainer: Sam moved an existing app into the new Codex desktop app on GPT-5.2 Codex extra-high reasoning to hunt bugs.
Both, in sequence
The candidate highlight reel builder and the Offsiteio estimator both started in Claude Code. Sam then moved them to Codex: the reel builder was "mainly built in codex" with Remotion, and the estimator went to Codex "after some different neat bugs and updates" (Episode 38, at 07:59). The estimator is now in daily production use.
Head-to-head
Jordan put the quality gap plainly in Episode 35:
One thing I would notice like repetitively with with Claude Code is that it would make mistakes, but be confident that there were no mistakes.
— Jordan Metzner, Episode 35
But the output of the code, both with 5.3 and now especially with 5.4 on extra high, I found to be significantly better quality.
— Jordan Metzner, Episode 35
Where each one fell short
Claude Code. The confident-but-buggy pattern above was the main reason Jordan moved. Anand Chandrasekaran of Arya Health argued in Episode 24 that Claude Code is "already very expensive when compared to everything else." In Episode 51, David Petrou pointed out that Anthropic makes subscription users go through Claude Code rather than other tools, while OpenAI has been "much more welcoming."
Codex. It is slow on the high-reasoning settings, and Jordan admits it is less enjoyable to use. Cost visibility was the bigger problem. In Episode 45, Jordan said the Codex computer use feature "definitely killed us on spend on tokens," and it took about two and a half weeks to find the cause. Sam added that the charges were spread across different consoles and coded differently.
Both. Ryan Aytay of Code Metal said code from Claude Code, Codex and their peers is great for apps but "isn't quite safe enough" on its own for mission-critical systems (Episode 48, at 06:21).
What our guests use
- Iliya Valchanov, Juma: the whole team moved from Cursor to Claude Code and canceled Cursor. They test Codex now and then but see no reason to move (Iliya's page, Episode 35 at 16:26).
- Ben Lerner, Espresso AI: pays for Claude Code and uses it for front-end work, while dashboards and throwaway pandas scripts come from Codex prompts. Neither is ready for serious backend infrastructure, he says.
- Mitesh Agrawal, Positron: "all of us are in love with Claude Code" (Episode 37).
Inside our own team, Matias Singer, a TPM who had started building with AI only two months earlier, used both Claude and Codex to build RADAR, which turns meetings into Jira tickets (Episode 39, at 01:09). He is a good example of what non-engineers can do with either tool.
Which one we reach for now
Our pick has flipped more than once, so dates matter:
- September 2025: Jordan's engineers moved from Cursor to Claude Code to Codex in about 12 weeks. OpenAI was subsidizing it, and Jordan said "we got no loyalty."
- January 2026: Opus 4.5 pulled us back. Jordan said his developers were "enthralled" with Claude Code (Episode 26), and in Episode 29 he said Ryz Labs was shifting more spend to Anthropic.
- February to March 2026: Sam found Codex extra high great for polish (Episode 30). By Episode 35, Jordan had moved back to Codex for code quality.
- June 2026 onward: Codex is Jordan's default.
My daily driver is usually Codex 5.5 extra high, kinda like turn it on there.
— Jordan Metzner, Episode 46
Jordan still lets each developer pick their own tool. Some Ryz Labs developers still prefer Claude Code. Switching is not that costly, but people get used to working one way. Jordan expects background agents that finish pull requests on their own to eventually pull the team into one ecosystem (Episode 35, at 18:10).
As of October 2026 (Episode 53), Jordan still uses Codex a lot. He also said Anthropic's newest Opus model was performing well and costs much less than Fable. Our advice is the same as Sam's in Episode 30: use the fast tool while you explore, then switch to the deep-reasoning one when you polish. And re-test every few months, because the leader keeps changing.
FAQ
Is Codex better than Claude Code?
For code quality, Jordan says yes: from Codex 5.3 onward on extra-high reasoning he found the output significantly better, and by June 2026 Codex 5.5 extra high was his daily driver. Claude Code is faster and more enjoyable, so Sam still splits his work about 50/50.
Which is faster, Claude Code or Codex?
Claude Code. Sam found it more instantaneous for quick iteration, while a Codex run on extra-high reasoning usually took five to ten minutes in his rep-by-rep trainer build.
Is Claude Code or Codex cheaper?
We never ran a pricing test. In September 2025 Jordan's teams maxed out Codex because OpenAI was subsidizing it, but in June 2026 Codex's computer use feature drove a token spend spike that took about two and a half weeks to trace.
Can you use Claude Code and Codex together?
Yes, and that is how Sam builds now: he started the candidate highlight reel builder and the Offsiteio estimator in Claude Code, then moved them to Codex for bug fixes and updates.
Do developers prefer Claude Code or Codex?
It splits. Juma canceled Cursor and moved its whole team to Claude Code with no plans to switch, while Jordan and some Ryz Labs developers moved to Codex for code quality and others still prefer Claude Code.