Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

If you've done any of Anthropic's AI courses, one message comes up again and again: use these tools ethically, and be transparent about when you've used them. I agree with that completely. But the more I lean on coding agents day to day, the more I bump into a question the courses don't fully answer — when an agent and I build something together, who actually gets the credit?

The conflict

There's a real temptation to take all the credit. The agent did the typing, but it was my idea, my review, my call on what shipped — so it's tempting to just put my name on everything and move on. The honest answer is messier than that, and pretending otherwise is exactly the kind of thing the courses warn you about.

The opposite problem is just as annoying. When I ask the coding agent to write my git commits, it cheerfully credits itself — even on commits where it didn't write the code at all. It bolts a Co-Authored-By line onto work it had no hand in. So I get inflated credit when I write the message, and the agent gets inflated credit when it does. Neither is transparent.

And the commit messages are enormous

On top of the credit problem, agent-written commits tend to be huge. Paragraphs of explanation, bullet lists, a co-author footer — for a one-line change. I like atomic commits with a subject line under 80 characters. The history should read like a changelog, not an essay.

A terser way to mark authorship

An idea I have come up with is using two emoji as shorthand: 👤 for me, 🤖 for the agent. It's tiny, it sits at the end of the subject line, and it tells you at a glance where the work came from without a five-line footer. Transparent and terse at the same time.

Human only:

ABC-1234: <action taken> 👤

Agent only:

ABC-1234: <action taken> 🤖

Collaboration:

ABC-1234: <action taken> 👤&🤖

The nice part is that this is enforceable. Rather than reminding the agent every time, I can write a skill and a set of instructions that bake the rule in: keep the subject under 80 characters, no long body unless I ask for one, and tag authorship honestly with the right emoji. The agent only claims credit for work it actually did.

How I write these posts (including this one)

It would be a bit rich to write about transparency and then hide how this post was made, so here's the process:

• I write a one-page outline. I create the post with a title and excerpt.

• I ask Claude to take my outline and write out the full post.

• I rewrite it afterwards so the post actually agrees with my opinions.

• I add photos or screenshots in later.

The advantages are obvious. Claude is much better than me at keeping things short and to the point, and it never makes spelling mistakes.

The disadvantages are the whole reason I'm writing this. The posts never truly sound like me, and the points often come out more generic and too polished — sanded down until any edge I had is gone. Which is exactly why this draft gets the 👤&🤖 treatment, and why I'll be rewriting the next paragraph in my own words before anyone reads it.

Spot the difference

See a lower case letter eg i instead of I, spot a spelling mistake, or found yourself double backing on a sentence: i probably re wrote that part.

The way forward

My conclusion is that we absolutely should use AI. We should just be honest about how we are using it.