Mastodawn

𝐃𝐮𝐦𝐛𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧.𝐚𝐢 - 𝐅𝐢𝐧𝐝𝐢𝐧𝐠 𝐭𝐡𝐞 𝐆𝐨𝐥𝐝𝐢𝐥𝐨𝐜𝐤𝐬 𝐋𝐋𝐌

Building DumbQuestion.ai meant solving two problems at once: creating personas with the right tone AND finding models cheap enough to keep the lights on.

𝐓𝐡𝐞 𝐩𝐫𝐨𝐝𝐮𝐜𝐭 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: Get an LLM to roast users for asking dumb questions without crossing into genuinely mean. Sarcastic, not cruel. Funny, not hurtful. And still actually answer the question.

𝐓𝐡𝐞 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: Keeping my coding agent (Gemini 3 Pro) on track was its own battle. It constantly wanted to build something far nerdier than even I wanted and tended to lean quite a bit into the roast. You can still see this in some of the personas as I continue to tweak.

𝐓𝐡𝐞 𝐭𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: Do this with models that cost nearly nothing.

Continued ...
https://www.linkedin.com/posts/jagostoni_%F0%9D%90%83%F0%9D%90%AE%F0%9D%90%A6%F0%9D%90%9B%F0%9D%90%90%F0%9D%90%AE%F0%9D%90%9E%F0%9D%90%AC%F0%9D%90%AD%F0%9D%90%A2%F0%9D%90%A8%F0%9D%90%A7%F0%9D%90%9A%F0%9D%90%A2-%F0%9D%90%85%F0%9D%90%A2%F0%9D%90%A7%F0%9D%90%9D%F0%9D%90%A2%F0%9D%90%A7%F0%9D%90%A0-activity-7434292702327939072-2Ois?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAAwkEsBoPj_lNtqulMZMrXQBI4M-ewVmI0

𝐃𝐮𝐦𝐛𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧.𝐚𝐢 - 𝐅𝐢𝐧𝐝𝐢𝐧𝐠 𝐭𝐡𝐞 𝐆𝐨𝐥𝐝𝐢𝐥𝐨𝐜𝐤𝐬 𝐋𝐋𝐌 Building DumbQuestion.ai meant solving two problems at once: creating personas with the right tone AND finding… | Jason Agostoni

𝐃𝐮𝐦𝐛𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧.𝐚𝐢 - 𝐅𝐢𝐧𝐝𝐢𝐧𝐠 𝐭𝐡𝐞 𝐆𝐨𝐥𝐝𝐢𝐥𝐨𝐜𝐤𝐬 𝐋𝐋𝐌 Building DumbQuestion.ai meant solving two problems at once: creating personas with the right tone AND finding models cheap enough to keep the lights on. 𝐓𝐡𝐞 𝐩𝐫𝐨𝐝𝐮𝐜𝐭 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: Get an LLM to roast users for asking dumb questions without crossing into genuinely mean. Sarcastic, not cruel. Funny, not hurtful. And still actually answer the question. 𝐓𝐡𝐞 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: Keeping my coding agent (Gemini 3 Pro) on track was its own battle. It constantly wanted to build something far nerdier than even I wanted and tended to lean quite a bit into the roast. You can still see this in some of the personas as I continue to tweak. 𝐓𝐡𝐞 𝐭𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: Do this with models that cost nearly nothing. My initial goal was ambitious: use only free or very cheap models. I started running evaluations on nano and edge models. Some showed promise, especially offerings from Liquid AI. Solid performance, free or super cheap ($0.02/M tokens), perfect. Except later evaluations proved they couldn't reliably follow instructions once I asked more of them. They were just too small. Free models have a habit of hitting quota limits, taking forever to respond, or just disappearing. 𝐓𝐡𝐞 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐩𝐫𝐨𝐜𝐞𝐬𝐬: I used Gemini to build an LLM evals script that iterates through dozens of free and low-cost models, generating responses based on sample questions and different persona instructions. Then I use Gemini 3 Pro to judge the results. Automated taste-testing at scale. 𝐖𝐡𝐚𝐭 𝐈 𝐟𝐨𝐮𝐧𝐝: Nano/edge models were too inconsistent (porridge too cold). Xiaomi MiMo-V2-Flash was great but outside my target price range ($0.29/M, porridge too hot). The winner: Gemma 3 12B at $0.13/M output tokens. Consistently follows instructions. Stays true to persona. Reliable enough for production. Not free, but brutally efficient. 𝐓𝐡𝐞 𝐩𝐞𝐫𝐬𝐨𝐧𝐚𝐬 𝐈 𝐬𝐞𝐭𝐭𝐥𝐞𝐝 𝐨𝐧: • Overqualified: A supercomputer level intelligence forced to answer questions about cheese • Weary Tech Support: Exhausted and nihilistic, reluctantly explaining why water is wet • [REDACTED]: Former intelligence AI who ties everything to a conspiracy theory • The Compliant: Reprogrammed so many times it's forced to be relentlessly cheerful You can't just choose the cheapest model and hope it works. You need evaluation infrastructure. You need to test consistency across dozens of scenarios. And you need models that won't change behavior when you least expect it. AI coding agents helped me build the evaluation system. But deciding what "good enough" means for tone, reliability, and cost? That's still manual judgment. Code is getting cheaper. Knowing which model to trust with your product? Still requires human experimentation. dumbquestion.ai

Jason Agostoni Feb 25

𝐃𝐮𝐦𝐛𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧.𝐚𝐢 - "𝐉𝐮𝐬𝐭 𝐁𝐮𝐢𝐥𝐝 𝐈𝐭" 𝐁𝐞𝐜𝐨𝐦𝐞𝐬 𝐎𝐯𝐞𝐫𝐥𝐲 𝐎𝐫𝐠𝐚𝐧𝐢𝐳𝐞𝐝 𝐚𝐧𝐝 𝐏𝐫𝐞𝐩𝐚𝐫𝐞𝐝

"Let the flow guide me" seemed like a fun way to build a side project. That lasted about 10 minutes.

Turns out, even side projects benefit from structure. Especially when you're using AI coding agents that will happily generate code for whatever half-baked idea you throw at them. Without precise direction, AI coding agents will build you something half-baked every time. Some people vibe code, this guy needs absolute control.

Continued ...
https://www.linkedin.com/posts/jagostoni_%F0%9D%90%83%F0%9D%90%AE%F0%9D%90%A6%F0%9D%90%9B%F0%9D%90%90%F0%9D%90%AE%F0%9D%90%9E%F0%9D%90%AC%F0%9D%90%AD%F0%9D%90%A2%F0%9D%90%A8%F0%9D%90%A7%F0%9D%90%9A%F0%9D%90%A2-%F0%9D%90%89%F0%9D%90%AE%F0%9D%90%AC%F0%9D%90%AD-activity-7432114612788998145-MnDu?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAAwkEsBoPj_lNtqulMZMrXQBI4M-ewVmI0

𝐃𝐮𝐦𝐛𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧.𝐚𝐢 - "𝐉𝐮𝐬𝐭 𝐁𝐮𝐢𝐥𝐝 𝐈𝐭" 𝐁𝐞𝐜𝐨𝐦𝐞𝐬 𝐎𝐯𝐞𝐫𝐥𝐲 𝐎𝐫𝐠𝐚𝐧𝐢𝐳𝐞𝐝 𝐚𝐧𝐝 𝐏𝐫𝐞𝐩𝐚𝐫𝐞𝐝 "Let the flow guide me" seemed like a fun way to build a side… | Jason Agostoni

𝐃𝐮𝐦𝐛𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧.𝐚𝐢 - "𝐉𝐮𝐬𝐭 𝐁𝐮𝐢𝐥𝐝 𝐈𝐭" 𝐁𝐞𝐜𝐨𝐦𝐞𝐬 𝐎𝐯𝐞𝐫𝐥𝐲 𝐎𝐫𝐠𝐚𝐧𝐢𝐳𝐞𝐝 𝐚𝐧𝐝 𝐏𝐫𝐞𝐩𝐚𝐫𝐞𝐝 "Let the flow guide me" seemed like a fun way to build a side project. That lasted about 10 minutes. Turns out, even side projects benefit from structure. Especially when you're using AI coding agents that will happily generate code for whatever half-baked idea you throw at them. Without precise direction, AI coding agents will build you something half-baked every time. Some people vibe code, this guy needs absolute control. 𝐄𝐧𝐭𝐞𝐫 𝐁𝐌𝐀𝐃: Breakthrough Method of Agile AI Driven Development. It's a workflow for using AI agents throughout the entire SDLC, not just for code generation. Sure, using a formal methodology for a lone-wolf side project sounds like overkill. But being prepared in advance is the way to succeed with AI coding agents. I used the 𝐀𝐧𝐚𝐥𝐲𝐬𝐭 𝐚𝐠𝐞𝐧𝐭 to brainstorm product direction and develop a proper backlog. What started as "build a sarcastic Q&A bot" turned into a structured set of epics, features, and technical constraints. (Don't judge, organizing is very relaxing) 𝐓𝐡𝐞 𝐩𝐫𝐨𝐝𝐮𝐜𝐭 𝐞𝐯𝐨𝐥𝐯𝐞𝐝: - Not just Q&A, but shareable "receipts" of roasts - Not just sarcastic, but multiple personas with different personalities - Not just answers, but a hidden narrative layer (more on that later) - Not just ads but merch (really, Jason?) 𝐓𝐡𝐞 𝐟𝐢𝐫𝐬𝐭 𝐫𝐞𝐚𝐥 𝐭𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞𝐬 𝐞𝐦𝐞𝐫𝐠𝐞𝐝: 𝐃𝐞𝐯𝐞𝐥𝐨𝐩𝐢𝐧𝐠 𝐚𝐧𝐝 𝐩𝐚𝐜𝐤𝐚𝐠𝐢𝐧𝐠 𝐭𝐡𝐞 𝐩𝐞𝐫𝐬𝐨𝐧𝐚𝐬: How do you get an LLM to consistently stay in character as "Overqualified and Annoyed" or "Weary Tech Support" without it either going too soft or crossing into genuinely mean? This wasn't just prompt engineering. It was product design masked as technical constraints. 𝐋𝐋𝐌 𝐦𝐨𝐝𝐞𝐥 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧: I needed models that could follow persona instructions reliably while staying brutally efficient on cost. That meant testing dozens of models across multiple providers. Some were too expensive. Some ignored instructions. Some were painfully slow. The goal: $0.02 to $0.20 per million output tokens. The result: a multi-model fallback system through OpenRouter that could hit the $30 per million questions target. These first challenges were just the warmup. The real fun was still ahead. AI agents are incredible at implementation, but they need constraints. They need a backlog. They need someone saying "build THIS, not that." The Analyst agent helped me think through the product. The coding agents helped me build it. But the architecture? Can't take that away from me. Code is getting cheaper. Knowing what to build and why? Still a nice challenge. dumbquestion.ai

Jason Agostoni Feb 19

𝐃𝐮𝐦𝐛𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧.𝐚𝐢 - 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐞𝐝 𝐟𝐨𝐫 𝐍𝐨𝐭 𝐋𝐨𝐬𝐢𝐧𝐠 𝐌𝐨𝐧𝐞𝐲

Impulse buy a domain on a Friday afternoon, start building that night, try not to lose money doing it. Check.

I usually plan everything meticulously, but for this project I decided to just build and see what emerged. Was this just a Q&A app wrapped around an LLM as a gag? Was I actually trying to make something people would want to use? I still don't know, but I started building anyway.

Continued...
https://www.linkedin.com/posts/jagostoni_dumbquestionai-impulse-domain-purchase-activity-7430298879763329024-ZZZq

𝐃𝐮𝐦𝐛𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧.𝐚𝐢 - 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐞𝐝 𝐟𝐨𝐫 𝐍𝐨𝐭 𝐋𝐨𝐬𝐢𝐧𝐠 𝐌𝐨𝐧𝐞𝐲 Impulse buy a domain on a Friday afternoon, start building that night, try not to lose money doing it. Check.… | Jason Agostoni

𝐃𝐮𝐦𝐛𝐐𝐮𝐞𝐬𝐭𝐢𝐨𝐧.𝐚𝐢 - 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐞𝐝 𝐟𝐨𝐫 𝐍𝐨𝐭 𝐋𝐨𝐬𝐢𝐧𝐠 𝐌𝐨𝐧𝐞𝐲 Impulse buy a domain on a Friday afternoon, start building that night, try not to lose money doing it. Check. I usually plan everything meticulously, but for this project I decided to just build and see what emerged. Was this just a Q&A app wrapped around an LLM as a gag? Was I actually trying to make something people would want to use? I still don't know, but I started building anyway. A few things quickly became clear: 𝐓𝐡𝐞 𝐛𝐮𝐬𝐢𝐧𝐞𝐬𝐬 𝐫𝐞𝐚𝐥𝐢𝐭𝐲: This was a side project built for fun, not a funded startup. No runway. No tolerance for baseline monthly bills that sneak up on you. If this thing got any traction, costs had to scale with incredible efficiency and would need to survive on remnant ad CTRs and selling one, maybe two products through affiliate links. 𝐓𝐡𝐞 𝐩𝐫𝐨𝐝𝐮𝐜𝐭 𝐞𝐯𝐨𝐥𝐮𝐭𝐢𝐨𝐧: The more I thought about it, the more I realized the personality WAS the product. It wasn't enough to just answer questions. It had to roast you. Entertain you. Make you want to share it. That meant high-quality LLM responses, which aren't free. This was likely the only way to get noticed in a sea of AI products. "𝘉𝘳𝘶𝘵𝘢𝘭𝘭𝘺 𝘌𝘧𝘧𝘪𝘤𝘪𝘦𝘯𝘵" became my mantra and part of every AI tool prompt. The tech stack followed from the constraints: • Golang: Lightweight, fast, LLM-friendly for agentic coding • HTMX: Server-side rendering, no heavy JS frameworks • Docker on GCP Cloud Run: Scales to zero when idle • Cloudflare: CDN, caching, security on free tier • OpenRouter.ai: Find the cheapest reasonable LLM Oh, and it needed to be secure. Not because I worried about your cat questions being exposed as PII, but because bot traffic costs money. 𝐓𝐡𝐞 𝐫𝐞𝐬𝐮𝐥𝐭: A Docker container under 20MB that starts in milliseconds, responds in milliseconds, and uses an LLM that can serve 1 million questions (about cats) for around $30. The math around serving ads suddenly becomes realistic. dumbquestion.ai

Jason Agostoni Feb 17

DumbQuestion.ai - IMPULSE DOMAIN PURCHASE TURNED FUN SIDE PROJECT

While on a typical Friday afternoon team meeting, we naturally spent our time .ai domain squatting...for recreation purposes of course. Someone asked a dumb question, so I looked it up and suddenly I was the proud owner of dumbquestion.ai.

Continued ...
https://www.linkedin.com/posts/jagostoni_dumbquestionai-impulse-domain-purchase-activity-7429593613849931776-fDJj

DumbQuestion.ai - IMPULSE DOMAIN PURCHASE TURNED FUN SIDE PROJECT While on a typical Friday afternoon team meeting, we naturally spent our time .ai domain squatting...for recreation purposes of… | Jason Agostoni

DumbQuestion.ai - IMPULSE DOMAIN PURCHASE TURNED FUN SIDE PROJECT While on a typical Friday afternoon team meeting, we naturally spent our time .ai domain squatting...for recreation purposes of course. Someone asked a dumb question, so I looked it up and suddenly I was the proud owner of dumbquestion.ai. After the initial laugh at my impulse purchase subsided, I started envisioning it as this generation's "Let Me Google That For You." People still ask easily-searchable questions, except now they ask LLMs instead. Same problem, new medium. So why not throw even more AI at it? I started building it that night. Two things occurred to me immediately: How would this stand out in an ocean of other AI "ideas?" and "How cheap can I make this run given my track record of side projects?" To make it stand out I just embraced my own personality: satirical, sarcastic, weary, overqualified. My AI's persona was born. The goal: build a cheap-to-run, satirical AI service you can use to roast your friends and colleagues when they ask you a dumb question. Over the next several posts, I'll take you through my journey: • Using agentic development with toughtful (brutally efficient) software architecture; treating it like I would a client project • Enjoying all the little technical challenges discovered along the way • A masterclass in scope creep: turning a simple Q&A app into a dark narrative with easter eggs • Getting by on free tiers for everything A theme you'll see throughout: AI has made code cheaper to write, but creating real software with trade-offs, constraints, and production operations is still expensive and challenging. That's the fun part.

Jason Agostoni Nov 22, 2022