Accidentally in Code

To be brutally honest, this (making Claude my social media manager) all started because I felt inadequate about self-promotion. And being an engineer, my response to feeling bad about something is to ask what I can build and what data would make the feeling go away.

So at the start of the year I built social-brain — a CLI that pulls my content data off every platform once a week, formats it into a prompt, and lets Claude produce a report. Six months on, running it is one of my end-of-week rituals. Here’s what it taught me.

The evolution

At first I was really into the dashboard – charts, glanceable metrics. It looked very nice and gave me a feeling that I had accomplished something. But over time, as my Claude setup became much more Claude Code than Claude web, this got folded into that setup as one of the weekly checks. It runs without me, puts the report in GitHub, and I read it.

That’s because what I actually needed was closer to a lab notebook. I wanted a longitudinal view that connected to everything else – something that could make suggestions on my content guidelines or cadence.

The report is now structured markdown: what worked, what didn’t, cross-platform patterns, experiment verdicts, and proposed rule changes. I read it, argue with it, and confirm or kill each proposed change, writing those changes across the relevant skills. The graphs are a nice way to feel a vague sense of movement (or not). The text answers “what should I do differently?” and reminds me of things I’d encoded and then broke for some (or no) reason.

The weekly cadence is useful. It generates actual suggestions. If I think they’re wrong – AI loves to over-generalise – I push back. If I hate them, I don’t do them. But more often than not I take something from it that I can act on.

What it showed me

Some things that six months of data in one place has made clear.

Ups and downs balance out and the baseline still moves. Sometimes you write something quickly at the airport and it is surprisingly popular. Sometimes you sweat blood over something it feels like no-one bothers to read. If I’m evaluating my social media without data, my opinion is wildly skewed by however well my latest posts did – or didn’t – do. What’s encouraging: those ups and downs, taken over time, show meaningful growth.

Traffic shifts. One thing I was able to see was how the balance of driving traffic to DRI shifted as it evolved. Jean has a large email list, and at first that dominated, but over time it’s evened out.

LinkedIn is an amp for launches. We all complain about LinkedIn, but apparently people do read content there. It drives traffic to everything, and it drives sales of DRIYourCareer. It’s also one of the few places where self-promotion is acceptable.

Newsletters are retention, not acquisition. I don’t promote my newsletter much, probably because I personally hate email, so it’s no surprise that email subscribers is a slow-moving metric for me. But those subscribers have good engagement – which means it isn’t an acquisition channel, it’s a relationship-building one.

Building out new funnels. I’d been coaching on the side for years, but quietly. After leaving my job I finally had time to expand it, so in January I built a dedicated coaching page and, later, a specific Calendly event type for coaching intros. The numbers are still small month over month. Not everyone I meet with becomes a coaching client, and that’s fine – I always tell people I’m not in the business of capturing people. But it’s moving, and I can look at it over time rather than worrying on limited data.

Pretty graphs

The chart at the top shows volume up and to the right, with coaching intros and DRI course sales both climbing through the same window and both inflecting around May.

Social Media Manager Claude is always telling me to add attribution tracking. Having worked in privacy, I keep pushing back – I think this chart is enough. Most content attribution, even when you can tag it, isn’t really known. It shows the last engagement, not everything that led up to it. Tagging gives the illusion of a causal machine, but doesn’t show you the longer term – which is what I care about. The person who read my blog for ages and then told someone about DRI – there’s no tagging that, even if I wanted to, which I don’t. I’ve always just tried to make as much public as possible and trust the value shows up in time.

Self-promoting without the ick

Over time, the question has become straightforward: how do I promote my own work without feeling gross about it?

I think I’ve settled on an answer. Better data helps but it’s not the whole thing. I have to be okay with the promotion itself – believing in what I’m telling people about goes a long way. And I have to be able to see it’s overall working. Doing it while hating it and suspecting it’s pointless is unbearable. Doing it while okay with it, watching it move something real, is manageable. I’m never going to love it. But I can be adequate at it.

I think this is where a lot of engineers get the ick, because we’ve all watched marketing graphs that looked great while the actual business results did not. Dashboards climbing, money moving – just to ad companies, not back into the business overall. Monopoly but with real money.

That’s the game I don’t want to play. I’d rather pick a couple of things that are real – is this sustainable, did it land somewhere – and be honest that most of the rest is a set of things that convince me, partly through data and partly through emotionality, to keep trying.

If you want the code, social-brain is open and any feedback is welcome. The newsletter signup is around here somewhere too.