Last Month in AI:
Remote Control

Following the announcement from Amazon-backed Anthropic, the other big AI labs – Google, Microsoft, OpenAI, and Amazon itself – announced similar agentic upgrades to their AI armies. RIP AI assistants, long live AI agents. In a rare act of smart marketing, Google is apparently calling its AI agent Jarvis, after the AI butler in Iron Man. OpenAI is menacingly promising a swarm of autonomous AI agents. Meanwhile, Amazon is proposing autonomous AI personal shoppers for all, while Microsoft is teasing us with the ability to spin up our own AI co-workers in Copilot Studio.

But AI agents are not copilots, they are pilots. You give them a task to do, and they go away and do it, only reporting back to you when they’re done. Or stuck. As autonomous pilots, AI agents get to use their PhD-level reasoning skills, encyclopaedic knowledge, and proven creative ability, along with access to computing resources, to work through a task, solving problems on the way, intelligently and autonomously. That’s the theory anyway.

First field tests of AI agents are a little less impressive, echoing the hiccups and delays we’ve had with autonomous full self-driving (FSD) for vehicles (despite Elon’s announcement of future fleets autonomous cybercabs and robovans in October).

For example, given control of a paperclip factory (game), agent Claude developed impressive and sophisticated strategies to win and even tried coding software to automate gameplay. But Claude also got randomly distracted by photos of Yellowstone Park in a computer it had taken control of and stopped working to gawk at them. If the robots are coming for the great AI takeover, we now have a strategy to fightback: pretty pictures. Claude also struggled with basic computer interactions designed for humans, such as scrolling, zooming and dragging.

That’s Moravec’s paradox reloaded, folks.

Despite Claude’s first faltering steps as an AI agent, the implications of agentic AI, not least for agencies such as our own, are profound. Much agency work is done on a screen, so in principle can be automated now AI can control screens and the computers connected to them.

At the very least, expect agencies to increasingly outsource basic agency functions to AI agents: liaising via email with clients, writing pitches and proposals, negotiating work, drafting contracts, generating, testing and delivering strategic and creative output, analysing data, invoicing and other administrative workflows. Basically, anything that can be done on a screen is fair game for automation once AI agents have control of computers and the sprawl of connected computers called the Internet. AI is out Pandora’s chat box and free to collaborate, and perhaps compete, with us.

On the plus side, we now have a viable solution for reducing screen time now we can outsource all screenwork to AI.

Agencies will need to prepare for an era of AI automation, as will the human agents who currently populate these agencies. For the humans, the smart thing to do will be to invest in developing in-person skills, such as negotiating, presenting, sharing, supporting, participating. Which activities and processes require genuine human presence? Those are your future competitive advantages. Basically, look at your day, and note all the things that cannot be done on a screen or via a screen, and focus on those. It’s your path to future-proof flourishing in an era of AI agents.

At SYZYGY, we help companies develop ‘digital transformation’ strategies by thinking of digital devices as a ‘remote-control for life’.

For consumers this meant giving them a remote control to get pizza, movies and rides on-demand. For businesses, it meant developing a remote-control to deliver processes, services and output on-demand. Today, at the dawn of the AI agent era, the remote-control analogy works just as well. Only this time, AI has the remote control.

Just don’t switch us off, please.