Why multiplayer AI has been out of reach

As cross-functional fire starters, there comes an inventible time when anyone in an AI-native bizops role needs to think about how to make these tools multiplayer.

This often starts with a few AI education classes, potential some ‘levels’ around AI use, some shared skills packets and artifacts, and maybe even a shared MCP or slack-based agent to communicate with.

In all of these cases, at least for me, there comes a point where multiplayer use of AI becomes next to impossible.

This is due to a few current realities

• The divide between those who know who to use AI and those who are semi-resistant is wide, and continuously grows as the tools themselves evovle rapidly
• Models and their respective harnesses are getting more technical as they fight for market share for engineers who are actively deploying these solutions at scale
• Outside of the skills and artifacts and various slack bots, there’s been very little attempt to make these tools multiplayer.

The reality is that when it comes to broad-based intelligence it’s extremely difficult to put all the use cases in a simple box.

This has long been why workflow based SaaS tools exist and also why service based businesses (agencies) thrive off the ‘complexity’ of deploying these solutions across organizations.

AI is and will be no different, at least in this preliminary stage.

So what can we do a responsible and forward thinking agentic athletes? Build in your own multiplayer system.

• Set the right ‘training wheels’. Instead of just making the skill or taking on the full automation of someone else’s responsibility, it's helpful to set up the right environment for someone to experiment with. For example, when a sales team member asked for specific partner analytics, I did a quick scoping excercise of what story he was attempting to tell, then I had an engineer set up a specific repo connected to Snowflake with the needed data. From here, I built v1 interface on Vercel that could easily be manipulated through any LLM based tool. For the new to AI salesperson, this meant simple natural language requests with immediate access to validated results.
• Allow for ‘controlled personalization’. The blessing and curse of AI is how much individual choice can impact the end outcome. Yet, similar to the training wheels, the end outcome doesn’t have to be based on some probabilistic, one-shot prompt. This is where that contained environment can also come in, e.g. hidden data infrastructure and hidden skills that keep answers as deterministic as possible. The only way I've found non-believers to start believing is when consistency isn’t sacrificed as they put their own touch to the processes and outputs. And there’s no greater way to discourage a fellow team member who's trying to learn these tools than to call their outputs AI slop (which inevitability they will be without these hidden guardrails in place).
• Think to the future of these tools. While these models continue to be deeply technical, there will be a day where there will be total obfuscation. Grok bots are a precursor to this. There’s no real ‘need’ for curating .md and sitting up thinking about context. There are lowering barriers to entry on native integrations. The chat interfaces are feeling less robotic. Sure, even Grok Bot is still missing out on the defined multi-player use but that’s inevitable. Importantly, if we’re forcing team members to create skills and similar technical relates concepts, we’re building too much for the now and not thinking about all the creative ways those concepts will fade away.

At least in my experience, ultimately the most helpful tool continues to be curiosity. Not taking outputs for granted and paying attention to the continual frictions these tools introduce, especially as model variations and tools evolve, is what makes other also feel ‘human’ in what is a very early process of re-imagining and vastly improving how we work.

The reason why I also decided to pick up writing again without the use of AI is to re-develop the balance of critical thinking mixed with a growth mindset that change is inveitable.

I strongly believe that agentic athlete role will be one of the most critical in this next phase, and I’d like to think that I’m not alone in feeling this way.