10x Tokens, 1x Output, 0x Thinking
Somewhere, a dashboard is very proud of you.
Something changed over the past few months.
I’ve been using AI tools for awhile and so have my co-workers. So, I’m used to seeing AI generated content at work, but lately it feels like it’s been getting worse.
I suspect it’s the spam of unfiltered slop. And I’m not referring to AI generated code. It’s everything else that has become AI integrated… the emails, Slack messages, useless meeting notes, and incoherent documents.
Here’s a sample of a few agentic artifacts I received recently:
- Tickets for problems that do not exist and provide recommended steps that do not apply to the non-existent problem.
- Code reviews for a fix in the wrong package, because another team’s agent confused our team’s wiki doc for theirs.
- Documentation five times longer than it needs to be because the author didn’t bother to give it a read-through before publishing it.
The Good Parts (Abridged)
To be sure, AI agents have made parts of my job easier.
I’ve offloaded routine work to scripts and bots: alarm investigations, sprint planning, and avoiding the death of a thousand OpEx tasks.
I’ve built some of my own tooling and used tools developed by other teams to address both complex problems and boring problems alike.
And I’ll just leave it at that, because this isn’t that kind of post.
Token Envy
Out of curiosity, I compared my token usage with some of my colleagues and I discovered a few of them were burning through 10x over the baseline.
Their output compared to the rest of the team? Roughly the same in quantity and impact. I’ve talked with high performers on both ends of the token spectrum and I’m convinced this metric doesn’t tell you anything.
What are these token whales doing that require their agents to run for hours at a time? And what happens to that output? Does anybody read it?
Meanwhile this metric has now become a goal, whether people are willing to admit it or not. Either way, other developers have told me they’ve been pinged at some point in the past year to “use AI more,” because some dashboard said so.
The result is that a tsunami of generated content arrives completely unchecked from their agent directly to my inbox.
Ctrl+C, Ctrl+V
C’mon, we’re all using the same tools. Don’t just copy and paste your agent’s response to me without even bothering to read it.
I recognize the bolded headers. The “Key Takeaways” section. The em dashes. The emojis.
It doesn’t matter to me that you’re using AI for help. But it’s incredibly insulting that you didn’t even bother to review it’s output to make sure it’s relevant or even accurate before sending it to me.
Key Takeaways
Your agent can write it — you still have to own it.
People have started to offload the one part of the job they were supposed to keep: critical thinking.
If you use an AI agent to help you, please take a minute to understand the problem you’re trying to solve and validate what your agent is doing before throwing your slop over the fence and making it my responsibility.
When you’re working with it interactively:
- ✅ Know the problem. If you can’t explain it, how will you know that you’ve fixed it?
- ✅ Check its sources. That wiki doc it referenced is from 2019.
- ✅ Cut the filler. Nobody needs five paragraphs of unnecessary context. Best practices still apply to design docs, proposals, etc.
When it’s running on its own:
- ✅ Define done. Verify tests pass, well-defined success criteria is met, and success metrics are tracked.
- ✅ Gate the handoff. Agents draft, humans ship and own it.
- ✅ Get a second opinion. Have another agent review the output. More tokens. You’re welcome, leadership.
My Agent Will Call Your Agent
Maybe the solution is to just let the agents translate all the slop for us.
Your agent sends me a message. My agent reads it, summarizes it, and drafts a reply. Your agent summarizes my reply. I don’t have to read anything. Token usage goes up for both of us. Leadership is happy.
I wonder what’s the AI slop version of “you have no idea what you’re doing, do you?”
My agent suggested the following:
“I just want to take a moment to express my heartfelt gratitude for you sharing this information! It’s truly a testament to your courage in sharing ideas at such an early stage. In today’s fast-paced landscape, alignment is more important than ever, so I’d love to dive deep into a holistic discussion to help clarify some of the foundational concepts and map out clear, actionable next steps, perhaps starting with the basics. Let’s unpack this together, step by step! 🚀
Happy to draft a very detailed agenda if that would be helpful!”