Darren Bounds

Building AI tools for my own business every day. Helping other companies do the same — hands-on, no fluff. Co-founder of The Playbooks | Vancouver

[Darren Bounds @darrenbounds]

One unexpected bonus: on my Mac, the Codex app already contained the CLI binary and was authenticated through my ChatGPT account.

I exposed that binary on PATH, so Claude Code could call it without another manual login or CLI installation.

[Darren Bounds @darrenbounds]

I removed one of the most annoying steps from my AI coding workflow.

Before:

Claude Code implements the change → gives me a Codex review prompt → I paste it into the Codex app → paste the findings back into Claude.

A manual round-trip for code reviews.

[Darren Bounds @darrenbounds]

The useful part is the separation:

- different model
- fresh review context
- read-only repository access
- triggered only at the review gate

Claude builds. Codex reviews. Claude verifies and fixes.

[Darren Bounds @darrenbounds]

@MatthewBerman @danshipper @every

[Darren Bounds @darrenbounds]

Works for specs, PRDs, outreach briefs, onboarding docs — anything where unclear thinking compounds downstream.

Draft with one model. Review with another.

[Darren Bounds @darrenbounds]

One LLM writes the ticket. A second one reviews it.

That extra step is where most of the real value is.

What comes back is usually the list of things that would've burned you in build.

[Darren Bounds @darrenbounds]

The loop I use:
1) Claude Code writes and owns the ticket
2) Claude Code generates a ready-to-paste Codex prompt (file path, context, specific questions included)
3) Codex critiques cold
4) Claude Code reviews the feedback and updates the ticket

[Darren Bounds @darrenbounds]

The skill strips:

scripts
styles
comments
framework payloads

…and keeps the text a visitor actually reads:

headings
body copy
CTAs
nav/footer text
image alt text

[Darren Bounds @darrenbounds]

Built a tiny agent skill because Claude Code/Codex were wasting huge amounts of context reviewing local HTML pages I was building.

So instead of feeding agents raw HTML, this extracts the readable copy first.

https://t.co/WD0Bq44Qs8

[Darren Bounds @darrenbounds]

Example: Shopify Canada homepage

Raw HTML:
~106.5k estimated tokens
Extracted readable copy:
~2.0k tokens

Result:
~52x smaller
~98% less context

[Darren Bounds @darrenbounds]

One caveat: the pre-built skills look US-focused. Tax prep references 1099-NEC and SE tax — not applicable outside the US.

[Darren Bounds @darrenbounds]

Anthropic just shipped a Claude for Small Business plugin that uses the QuickBooks tool

I've been running Intuit's official QuickBooks Online MCP server for our business accounting — through Claude Code and Codex the past few months.

https://t.co/hKfvzvyrcx

[Darren Bounds @darrenbounds]

The Claude for Small Business plugin takes a different approach: pre-built workflow skills — /close-month, /plan-payroll, /tax-prep — no developer setup needed.

Excited to see how much of the QBO MCP's depth it eventually incorporates.

[Darren Bounds @darrenbounds]

ChatGPT not holding back on where Codex fits in.

[Darren Bounds @darrenbounds]

@PudgyPop @pudgypenguins @LilPudgys @brave @guanocoin @AttentionToken @aingeth Who should I bet on hmmm?🧐

[Darren Bounds @darrenbounds]

Exciting to be a part of the Alimentiv and Satisfai team, helping drive innovation in IBD research.

[Alimentiv @alimentiv]

Alimentiv and Satisfai Health Announce the Commercial Availability of a Highly Innovative AI-Powered Scoring Tool for the Assessment of Ulcerative Colitis in Stage I through IV Clinical Trials. Click the link to access the full press release: https://t.co/HSlFsIT4t8