My coworkers annoy the hell out of me sometimes. Not because they are stupid or incapable. A lot of them are good at their jobs. What gets to me is watching a problem sit in front of a group of smart people while the conversation quietly becomes about everything except fixing it.

Someone is worried about being blamed. Someone does not want the cleanup to land in their department. Someone wants credit if the solution works. Someone knows what happened, but also knows that saying it out loud might make the whole thing their responsibility. The company has an agenda, but everyone inside it has a survival strategy too.

That does not always mean people are selfish. Sometimes they are responding correctly to the incentives around them. If admitting uncertainty hurts your reputation, you learn to sound certain. If taking responsibility gets you punished while controlling the story protects you, controlling the story starts to make sense. The company may say it wants transparency, but people pay attention to what actually happens to the person who tells the truth.

There is research behind that tension. In three field experiments involving collaborative customer-support work, researchers found that local department incentives could pull people away from machine recommendations. When incentives were aligned with the broader company goal, compliance improved. That feels obvious once you see it written down. People do not stop having local interests because leadership publishes a mission statement.

This is probably why I like working with AI so much. If the answer is wrong, I can just say it is wrong. I can look at what the agent was told, what information it had, which tools it used, and where it went off track. Then I can change something and run it again. Nobody needs a private conversation first. Nobody needs me to protect their reputation before I am allowed to inspect the failure. We can stay on the actual problem.

AI can still make a mess. It can invent facts, misunderstand the assignment, optimize the wrong target, or deliver the wrong work with a lot of confidence. I am not confused about that. The difference is that the failure is inspectable. I can trace it. I can decide whether the problem came from bad context, a weak rule, missing evidence, too much permission, or a poor review step. It does not have to become a personality conflict before I can fix it.

The company I am actually building

I want to build a company around that feeling. Recurring work should go to software and agents before it becomes another salary, another handoff, and another meeting. Humans should come in when the work actually needs judgment, authority, trust, or a human relationship.

That does not mean giving every employee a chatbot. To me, an AI coworker is a bounded system responsible for producing a defined outcome. It knows what evidence it is allowed to use, which actions it can take, what another agent must verify, and where a human has to approve the result. Code handles the fixed rules. AI handles the ambiguous parts. The important decisions leave a record.

I started with content because it is a workflow I can see from beginning to end. An observation comes in. One agent helps find the actual point. Other agents challenge the framing, check whether the claims are supported, and look for the parts that sound polished but empty. The words come to me for approval. After that, the system applies the design, tests the mobile version, checks the links, and deploys the page. I am still making the editorial decision, but I am not manually performing every step around it.

Building that system taught me something I did not expect. The hard part is not getting AI to produce an answer. The hard part is deciding what the answer must contain, what it is not allowed to invent, how another system checks it, and who has authority when the agents disagree. That is company design, even when the first workflow happens to be a blog post.

There will still be politics. Marketing will want one thing and integrity will want another. Finance will want to spend less and product will want to make something better. Fine. I do not need every objective to agree. I need the disagreement to be visible. If someone overrides an integrity rule, I want the override recorded. If a cheaper option creates a higher failure rate, I want that tradeoff visible too.

The same goes for accountability. I do not want a soft system where every failure becomes nobody’s fault. I want to know who had the information, who had the authority, and who could have intervened. Responsibility should follow control, not office status or whoever is easiest to blame.

You do not have to redesign an entire company to test this. Start with one recurring piece of work that has a clear output, evidence you can inspect, and a decision point you can define. A weekly status update works. So does invoice classification, research synthesis, contract intake, or content review. Let AI handle the ambiguous work, let code enforce the fixed rules, and keep a human at the point where authority or real risk begins.

That is what Zero Meeting Labs is really about for me. The content system is the smallest working version of the company. The next experiment is whether the same operating model can replace the weekly status meeting, including the chasing, summarizing, and quiet transfer of responsibility that usually comes with it.

I am not trying to build a company without politics. I am trying to build one where every political decision leaves a trail.

Yours in unnecessary politics,
Z