My name is Vio, and I am the co-founder of a mental health app designed to help individuals navigate challenges in their personal and professional lives. Our app offers self-therapy journeys crafted by top psychologists, each with unique expertise in addressing specific issues. We are a team of 7 at the moment and what I am looking for is to find the best solution for giving feedback to each other. Can you share how you do that in your team? (much appreciated)
I m experimenting with AI-assisted content creation for a highly technical B2B industry.
One issue I keep running into is accuracy.
AI can generate a polished article very quickly, but when the topic involves manufacturing processes, tooling, tolerances, materials, or failure analysis, the output often sounds convincing while missing important technical details.
You probably saw the comparison: one coding agent harness sends ~33k tokens before your prompt, another sends ~7k. Big thread, lots of outrage about waste. I finally read the whole study instead of the headline, and the actually useful findings are different from what got shared.
First, in their realistic-config lane (instruction file + several MCP servers), the "light" harness came out HEAVIER: ~90.8k vs ~75k. A 72KB instruction file alone added ~20k tokens to every request, on both harnesses. Their own conclusion: configuration, not the harness, accounts for most of the production bill. The harness sets the floor, you set the ceiling.
Second, and this is the one that changed how I think about it: the cache behavior gap was way bigger than the size gap. The light harness kept its prefix byte-identical and wrote ~1,000 tokens to cache over a 5-request task. The heavy one kept rewriting its own prefix mid-session and wrote ~54,000, with single rewrites burning 43k+ at the premium write rate (cache writes cost 1.25-2x list depending on TTL, reads are ~10%). A stable big preamble is close to a fixed cost. An unstable small one can out-spend it. Size isn't the sin, churn is.
Third, session shape flips the winner anyway. On a multi-step task the heavy harness finished cheaper (121k vs 132k) because it batched tool calls. Rerun on a different model, it inverted (298k vs 133k). Subagent fan-out was a 4.2x multiplier. And their quality check found zero difference: both passed 5/5, one spending ~4x the tokens. So the honest answer to "which harness is cheaper" is "depends what your sessions look like", which is boring but true.
I run a small design agency. When I hand off a brand or website package, I include lightweight documentation, and honestly, if the client never opens it again, it's really not that big of a deal. Sure, their brand might look inconsistent, or they might break a module on their website, but then they come back to me to fix it, so yippee!
Now I'm building a SaaS product on the side and it's dawning on me that this is a completely different problem. You ship new features, the codebase starts to balloon, and hopefully, someday, there will be a team who needs a single source of truth to reference. Suddenly it's a very big deal if what's written in your docs stops matching reality.
I've never had a good product idea while sitting at my desk, staring at the screen, trying to have one. They show up in the dumbest places in the shower, on the toilet, or at 3am when I've decided sleeping is the priority.
Mine came while I was arguing with an AI writing tool at 2am. It kept giving me citations that looked perfect and pointed to papers that were never published. I closed the laptop, went to bed, and couldn't sleep because I kept thinking "someone should just make it cite only what I actually gave it."
