Agents Go Mainstream (and Every Picture Learned to Draw Itself)
Manus put autonomous agents in the group chat, MCP became a de facto industry standard, and one image model turned the internet into Studio Ghibli. Creative work just joined the conversation.
Three seemingly separate stories this month, one underlying event.
Story one: a Chinese startup's agent product, Manus, became the tech world's obsession — an AI you give a goal, not a prompt, and it plans, browses, writes, and delivers while you sleep. Waitlist invites traded like concert tickets.
Story two: OpenAI adopted Anthropic's Model Context Protocol. When the two fiercest rivals in a technology race agree on the plug standard, the market is telling you the product category — agents wired into real systems — has arrived. The connector I flagged last November as "remember this name" is now the industry's shared nervous system.
Story three: GPT-4o's native image generation launched, and within seventy-two hours the entire internet was re-rendered in the style of Studio Ghibli. Charming, viral — and a genuinely uncomfortable moment for every illustrator watching a decade of craft reproduced as a filter.
The one underlying event
The frontier stopped being about answers. Agents mean the machine executes multi-step work; standardized connectors mean it executes inside your actual tools; native image generation means the execution now includes the visual and creative layer. In my Star Trek essays I wrote that machines handling labor could free humans for creativity and exploration. This was the month the machines politely pointed out that they'd like a share of the creativity too.
I hold two truths here. Creative output is being commoditized at shocking speed. Creative authorship — taste, meaning, lived perspective, a name that stands behind the work — is appreciating just as fast. The artists who thrive will sell the second. The ones crushed will be those paid piece-rates for the first.
Actionable Steps
Companies: your agent pilots should now assume standardized tool access — build the audit and permission layer before you scale the fleet, not after the first incident. Creators: put your name, your story, and your judgment forward; anonymous production work is the exposed flank. Everyone: run one real task through an agent this month, end to end. You need a calibrated gut, not a curated demo reel.
The machines learned to act this quarter. Next quarter, employers start acting on that.
Related reading elsewhere
Follow the experiment
One essay a month on where the post-labor economy is actually heading — capability, evidence, policy, and a live experiment in post-labor income. No hype, receipts included.
Follow Kenn on LinkedIn