In No AI Slop: A Manifesto, slop is work that takes the recipient more effort to understand than its creator put into making it useful. That’s a good test for anyone using AI at work. It takes seconds to generate pages of text. Someone still has to read them.

The speakers and attendees of AIE Code Summit 2026 set out five values for working with AI. You can use them when you’re deciding whether to open a chat and again when you’re ready to put your name to the work.

1. Outcomes over output

It’s easy to produce enough slides, pages and words to look as though you’ve done a lot of work. If the aim is to get your boss off your back, AI is very good at that. The manifesto asks a more useful question. What does the work actually achieve?

Before you ask AI to make anything, name the person who’ll use it and the decision or action it should help them take. That gives you a way to judge the result. If you can already tell them what they need in a couple of sentences, send those.

2. Understanding over speed

Here’s my test. If your computer died right now, could you explain what you’ve just produced? Could you talk someone through the main point, the evidence behind it and why you reached that conclusion?

The point where I’d slow down is when AI has produced so much that you haven’t read half of it. Work through one part at a time. Check the source behind an important claim. Put the reasoning into your own words before you move on. Your brain needs a chance to keep up.

3. Our own thinking over a model’s first draft

I’d leave AI out of the first draft when the point of the work is to say what you think. By the time the model hands you a draft, it’s already chosen an angle, an opening and which ideas deserve the most space. You can change every sentence and still be developing its argument.

Talking into your computer is an easy way to start. Say what you think, why you think it and where you’re still unsure. Let it come out in the wrong order. Then give AI the transcript and ask it to organise your points. You’ve given it something of your own to work with.

4. Better work over easier work

This is the big one for me. If the appeal of using AI is that I can skip the hard part, that’s a red flag. What interests me is the work I could do with it that I couldn’t do as well on my own.

Look for a difference you can point to in the finished piece. In your research, which source did AI help you find that changed your understanding? In your analysis, did it spot a gap in the reasoning that you’d missed? With a visual or a presentation, what can someone now grasp that was hard to get across in words? Those are good reasons to use AI.

5. Improving the system over patching the output

I used to patch a sloppy AI answer until it was just about good enough. The trouble is that editing the answer does very little for the next piece of work. When AI gets something wrong, look at what it had to work with.

  • Context. Did it have the brief, your reference material and an example of work you’re happy with? Put those in a project folder it can read.
  • Access. If the information it needed already lives in one of your work systems, a connector or plugin can let it consult the source.
  • Instructions. If it had the right material and still did the job badly, change the instructions for that task. A skill is a set of instructions the assistant can reuse.

You can dictate the correction while it’s fresh. Tell the assistant what it got wrong and how you want it done next time. If it can edit the skill, ask it to make the change and show you the saved wording. Otherwise, save the rule in the project instructions yourself. Try the task again in a fresh chat to see whether the correction stuck.

Five checks before you share AI-assisted work

Pick one piece of work you’re about to send. Tick the statements you can stand behind.

  1. Point 1
  2. Point 2
  3. Point 3
  4. Point 4
  5. Point 5

The manifesto also asks you to say when an AI agent drafted something you’re sharing. Tell the reader what you checked. If a question comes back, you should be able to answer it yourself.