I'm Masaki Hirokawa — an artist and solo app developer, and the person running Gemini Lab.
For about the past month, I've changed one small thing about how I work with Gemini. Before any decision that actually matters, I deliberately ask: "Please give me only counterarguments. I don't want any agreement."
To be honest, this has made a bigger difference than I expected. So today, instead of a how-to, I want to write something more reflective: how the way a solo developer makes decisions has quietly shifted as AI has moved into the workflow. This isn't a deep technical post — read it as a small observation from someone who builds and ships apps for a living.
Why default Gemini doesn't surface counterarguments
Honestly, Gemini tends to be an enthusiastic partner. That isn't a flaw — it is what conversational assistants naturally do when you don't direct them otherwise.
If I say, "I'm thinking of designing a daily push notification into my app," the response will almost always lean positive: "That's a good idea. To implement it, you could…" Nothing about that is wrong. But the question I actually wanted answered was, "Where are the holes in this decision?"
Solo development is a job where, in the end, the only person making the call is me. There is no internal review committee, no manager who can reject my pull request. Precisely because of that, I've come to believe that deliberately listening for the opposing voice — before deciding — is something I have to architect into the day.
The "counterargument prompt" I actually use
Concretely, I've been asking like this:
Regarding the idea I'm about to share — do not give me any agreement or supporting points. Surface only the counterarguments, concerns, and failure scenarios you can imagine. Prioritize the ones I might be missing.
Three things matter here.
First, I explicitly forbid agreement. Without that, Gemini will instinctively try to "balance" the response with both sides. Setting a forbidden move inside the prompt actually changes the shape of the output.
Second, I ask for the things I might be missing, not generic counterarguments. The aim is to get pushback that targets blind spots inside my own assumptions. Because Gemini already knows some context about me — solo developer, ad-revenue dependent, mostly Japanese-speaking users — it can produce objections that are actually shaped by that context.
Third, I ask two of them. I send the same prompt to Gemini 2.5 Pro and 3.1 Pro and look at the overlap and the gap. Whatever both flag is a "genuinely weak spot." Whatever only one flags is a "perspective worth keeping in mind."
Three decisions counterargument-mode actually saved this month
Nothing here is dramatic. But over the past month, there were a handful of moments where this mode visibly changed where I landed.
Case 1: Whether to ship a community feature
In one of my wellness apps, I had been considering a community feature. There were user requests for it, and my gut said it would lift engagement.
When I sent the idea to counterargument mode, this came back:
"For users who came to your app for wellness and calm, exposing them to other people's posts may erode the very experience they came for. A community feature trades engagement for the dilution of the app's core promise."
That was a perspective I had completely missed. The users who write in with feature requests are the engaged minority; the silent majority's experience may be a different shape entirely. In the end, I shelved the community feature and instead built a "private journal mode" — something a single user uses alone.
Case 2: When to raise the subscription price
I was looking at a price increase. Cost structure made it necessary.
The counterargument-mode reply was direct: "This is the moment when users are emotionally expecting AI features to expand. If you raise the price without making explicit what they are getting in return, this will read as 'just a price hike,' and churn will rise."
So I went ahead with the increase, but timed an AI feature update to ship the same week, and explicitly wrote in the release notes, "With this change, here is what you can now do." Churn came in lower than I'd modeled.
Case 3: How to position Gemini Lab's articles
To be honest, I had been wavering on the editorial direction of Gemini Lab. Should articles get longer and more comprehensive? Or shorter and more pointed?
Counterargument mode said: "Comprehensive articles are exactly what AI is starting to mass-produce. If you compete on coverage, Gemini and Claude can write ten times faster than you, and you will lose by attrition. The defensible angle for you is the resolution of what you, as a solo developer, have actually lived through."
I had vaguely felt this, but never put it into words. Since then, before writing a piece, I ask myself: "Does this article carry a resolution that only my lived experience can produce?"
When NOT to use counterargument mode
Having said all that, there are times you absolutely should not use this mode.
The first is early in the morning, or when you're tired. Being on the receiving end of nothing but pushback freezes your judgment. I once tried this on a low-energy morning and effectively didn't move forward with anything that day. Counterarguments from AI are something to receive when your inner ground is steady.
The second is when the idea is still a seed. If you spray counterarguments at an idea before it has taken any shape, the seed itself doesn't survive. I've learned to leave a "still figuring it out" period intact, and only switch to counterargument mode once a rough outline is visible.
The third is when a human conversation is the right call. If I have a co-developer, or if the decision touches family or a partner, talking to a person is faster and more accurate than asking AI. It's worth remembering that AI is convenient enough that we sometimes forget human conversations are even an option.
What "just argue against me" actually changed
If I had to put into one sentence what shifted after a month, it would be this.
I used to think of Gemini mainly as a tool to make work faster — coding, research, polishing prose. Since starting to use counterargument mode, I now also see Gemini as a tool to make decisions better.
For a solo developer, this is a bigger shift than it first sounds. For someone who has to make every call alone, having a built-in mechanism to take in a perspective other than their own, on their own responsibility, directly affects long-term quality.
To be clear, the counterarguments Gemini surfaces are not always right. Honestly, about half of them I read and decide, "That doesn't apply to my context." But a decision that has been examined through the lens of objection is visibly stronger than one I made alone from the start.
A small step you can try today
If reading this made you think "maybe I'll try it," you don't need to start with a big decision.
Tonight, before you go to bed, take a small thing you decided this week — "this is the title for my next blog post," "this is the next feature I'll build," "I'll subscribe to this service" — and ask Gemini, "Just give me the counterarguments."
You'll probably notice more holes than you expected. And about as often, you'll come away thinking, "No, I still want to do this," with more confidence than before. Either way, the decision walks out the door a little stronger than it walked in.
I'll keep this practice going through next month. If anything new comes from it, I'll write about it again.
Until then, take care.
If you'd like to pair this practice with deeper reading on cognitive blind spots, Thinking, Fast and Slow by Daniel Kahneman is a classic worth your time, and our Complete Guide to RSFC Structured Prompts for Gemini helps you build your own variations of the counterargument prompt.