The backlash against AI slop is right about the output and wrong about the machine. We hired AI into the one role it was never suited for.
There is a good argument going around that AI is producing a lot of bad work. It deserves to be taken seriously, because it is true.
Researchers at BetterUp Labs and the Stanford Social Media Lab, writing in Harvard Business Review, gave it a name: workslop. Output with the shape of finished work and none of the substance. Forty percent of US desk workers received some within a single month. Each instance cost the recipient close to two hours of rework, roughly $186 per employee per month. Forty-two percent trusted the sender less afterward. Outside the building, a 2026 DoubleVerify study of 22,000 consumers across 22 markets found 42% say low-quality AI content damages their opinion of the brand behind it.
Those numbers are real, and if you lead a brand you have felt them.
We gave it the wrong job
Look at what we actually asked the machine to do. Write the campaign. Have the idea. Name the product. Decide what the brand sounds like. We handed AI the creative director's seat and the strategist's seat, the two roles that run entirely on taste, context, and knowing what your CEO said in last quarter's board meeting.
Then we were surprised when the work came back plausible and hollow.
MIT's State of AI in Business 2025 found 95% of organizations investing in generative AI were seeing no measurable return on the P&L. Read that as a hiring story rather than a technology story and it is pretty damning.
So, is that it? Should we divest? I don't think so. My whole point is that the tool was handed a job description before anyone had discovered what it was actually extraordinary at. While that argument was playing out in public, something far more useful was happening in the background.
The orchestration layer grew up
The most consequential AI advancement of the past two years was more than a smarter model. It was the connection layer that lets a model reach into the systems where your business actually lives.
Model Context Protocol, and the agentic infrastructure built around it, changed what is possible. A platform like Claude can now work directly inside your CRM, your asset library, your product information system, your ticketing queue, your analytics, and your marketing automation, as one coordinated sequence, against your real data. Not describing what somebody ought to do next. Doing it.
That matters because of where enterprise time actually goes. Your senior marketing operations lead is not stuck for lack of ideas. They are stuck exporting from one platform, reconciling it against a second, correcting the taxonomy in a third, staging the changes, and chasing three approvals. Multiply that across an operation and your best people spend most of the week navigating interfaces instead of thinking.
That is the job AI is genuinely, unreasonably good at. Complex, multi-step, cross-platform, precision-dependent work that no one enjoys and every operation quietly runs on.
This is what maturity looks like
Every powerful machine we have built has been a supplement to a human vision rather than a substitute for one. The camera did not replace the eye. The synthesizer did not replace the composer. The printing press did not replace the author. In each case the machine absorbed the mechanical difficulty and handed the person back the part only a person can hold, which is the point of view.
The first era of enterprise AI got that exactly backwards. It tried to automate the point of view and left the mechanical difficulty sitting right where it was. Slop is what that inversion produces, reliably, every time.
What that looks like when we build it
At Trellist Marketing & Technology, AI orchestration is now a major of what we do. We build the custom MCP servers and agentic infrastructure that give your AI layer secure, governed access to the platforms your business runs on. Then we design the system around it: which agents exist, how work and data route between them, where the human checkpoints sit, and how the whole thing gets measured.
We forward-deploy the team that builds it, inside your business rather than back at ours. An orchestration engineer owns the architecture. Agent engineers build and harden the agents against your live data and workflows. A business deconstruction analyst takes the process apart so AI can take it on. And a creative lead sits on that same team, accountable for brand, voice, and experience quality at AI speed, because we are not willing to ship you a system that produces the thing this article opened with.
Two commitments we stand by. The first is that you will see impact as soon as the MPCs go live. Guaranteed. And the second, you are not doing this alone, which the evidence says matters: MIT found that AI capability built with a partner reached full deployment about twice as often as capability built internally from scratch, 66% against 33%.
AI deserves a new job.
Let's give it one.
Book a working session: trellist.com/contact-us | solutions@trellist.com | (302) 778-1300
Sources
- BetterUp Labs + Stanford Social Media Lab, Harvard Business Review, September 2025 — "AI-Generated 'Workslop' Is Destroying Productivity" (1,150 US desk workers)
- DoubleVerify global consumer study, July 2026 — 22,000 consumers across 22 markets
- MIT NANDA, State of AI in Business 2025: The GenAI Divide, July 2025 — 95% no measurable P&L return; 66% vs 33% partner-built vs internal deployment success
.jpg)
