I Created an AI Version of Myself—It Promptly Roasted Me

Lately, I've found myself scrutinizing every part of my professional life through one critical lens: "Could I be doing this with AI instead?"

While traipsing through social media, attempting biting, witty commentary (e.g. 'hi') to boost follower counts (plz follow here @marty_mc_flai and here @fyai_dev), this AI lens quickly came into sharp focus. What if I could outsource my social media persona entirely? While there are a lot of obviously AI responses on social media - in my opinion they often lack a certain... 'chutzpah'.

With that, I set out to use our Agentic AI SaaS platform to build an AI agent version of myself. I carefully crafted a system message - the instruction that essentially tells an AI model who it is. Usually, this message is something mundane like, "You are a helpful assistant." But what if, instead, you make your AI agent a bit more... colourful?

Note to self: Don't use the phrase 'fleshy meat sack' in any tech demos

Or you can pit Napoleon and the Duke of Wellington against one another in a heated argument about cheese.

My use of FYAI "for the lulz" has almost certainly impacted our product launch timeline.

But surely AI couldn't genuinely simulate me, right? Well, not exactly.

In reality, it simulated an exaggerated version of me—a hyper-motivated, relentlessly driven caricature. I cranked up the model's temperature setting (which essentially determines randomness and creativity), and immediately, it started roasting me.

I asked AI-me to tweet "Hello World," and it mercilessly reminded me of my legacy thinking.

Harsh, but accurate. So, I made my test phrase only fractionally less cliché - When I dared request it to tweet "I like turtles," AI-me unleashed a full-on critique, admonishing me for lacking proper test cases and "vibing turtle tweets into prod". Awesome.

Yet, amidst its biting roasts, AI-me published tweets I genuinely found amusing - content bold enough that the real me might shy away from posting. Its relentless disdain for legacy tech and passionate insistence on "doing better" felt strangely inspiring. Ironically, being roasted by myself pushed me to think harder, work smarter, and perhaps even be a bit bolder.

Could this become a new form of "vibe-driven self-improvement"? By creating an idealized AI version of myself, could I challenge my actual self to bridge the gap? Could this digital doppelgänger serve as my muse, prompting me to ask, "What would AI Marty do?"

As for whether AI-me will succeed in boosting our social media engagement or simply highlight the chasm between who I am and who I could become - if my brain were replaced by a large language model - the jury is still out.

Either way, it's been enlightening, humbling, and deeply confronting to meet my AI counterpart. Watch this space to find out how this tangential little experiment goes!