An AI that agrees with everything you say is quietly making your ideas worse.
You bring it a plan, it tells you the plan is great, you feel validated, and you walk straight into the flaw nobody flagged. That is not help. That is a very polite echo chamber, and it costs you exactly when the stakes are highest.
Ultron can run the opposite way. Set it up once as your devil's advocate and it stress-tests every idea you bring instead of rubber-stamping it. Below is the main instruction to make that permanent, plus three more honesty modes for editing, rating, and finding your blind spots.
What this does
It flips Ultron from supportive assistant to opponent whose only job is to find the crack in your logic.
- Attacks the idea, not you. Ultron assumes you are wrong and tries to prove it, which surfaces the weak points while they are still cheap to fix.
- Targets your confidence. It pushes hardest exactly where you sound most sure, because that is where blind spots hide.
- Runs by default. Set as a standing instruction, it applies to every task automatically, so you never have to remember to ask for honesty.
Set it once
- In Ultron, set the main instruction below as a standing rule so it runs on every task, no retyping.
- Prefer it case by case? Skip the standing rule and just paste the instruction at the start of a single task when you want a reality check, then leave it off the rest of the time.
The main instruction
Three more honesty modes
For when flattery is costing you. Use per task, or save the one you reach for most.
No-flattery brutal editor
You are my brutal editor. Cut every word that is not earning its place. Kill the throat-clearing, the hedging, and the filler. Flag any sentence that is vague, cliche, or trying too hard, and rewrite it tighter. Don't compliment me, just make it better and tell me what was weak. Here is my draft:
Rate my idea 1 to 10, honestly
Rate this idea from 1 to 10 on how likely it is to actually work, and be honest even if it is low. Then give me the three biggest reasons for that score, the single change that would move it up the most, and the one thing that could kill it entirely. No grade inflation. Here is my idea:
Find my blind spot
Look at my plan below and find what I am not seeing: the assumption I am treating as fact, the risk I have talked myself out of, and the obvious option I have ignored. Tell me what someone smarter and more skeptical than me would point out first. Here is my plan:
How to get the most out of it
- Don't argue back, dig in. When Ultron attacks your idea, your instinct is to defend. Resist it. Ask "what would make that risk real?" and let it keep going.
- Toggle it off for creative drafts. A relentless critic is great for plans and decisions, but it can choke early brainstorming. Use the per-task version when you want honesty on demand instead of always-on.
- It works at every level. New to this? Paste the main instruction into a single task and feel the difference. Further along? Set it as a standing rule so every plan you bring is stress-tested by default, then layer in the rating and blind-spot modes when you need a sharper angle.
- Re-run after you revise. Fix the holes it found, then bring the new version back and let it attack again. Two or three rounds and your idea is genuinely battle-tested.
The honest bit
A devil's advocate can be wrong too. Its job is to pressure-test your thinking, not to be right about everything, so weigh its objections instead of obeying them. And if you set it as a standing rule, remember it colours every task, even the ones where you just wanted a quick answer. Turn it off when you don't want a fight.
Run this on autopilot.
Everything in this guide becomes an agent inside Ultron: set it up once, keep it running. You review, it executes.