Are you guys adapting straight away or truly look into whether the new model suits you?🤔 https://www.youtube.com/watch?v=XKJCAusxfWA&t=174s

Two months ago when I joined the accelerator I thought LinkedIn was there for people to find a job. Fast forward to today, 4k followers and growing, and I’ve got people literally reaching out themselves to book discovery calls. All this achieved by simple methods I learned through this community and the AAAAccelerator. Super grateful and excited for what’s to come 🥂 keep in touch: https://www.linkedin.com/in/marios-dellis?utm_source=share_via&utm_content=profile&utm_medium=member_ios

Just over an hour on video, recorded, with his store admin open on screen. I went in with a list of numbers I wasn't allowed to leave without. I also went in with a favourite answer, and that turned out to be the problem. After discovery I was sure the bottleneck was content. He'd told me he had loads of footage that never got published, so I'd basically written the report in my head: a system to get the backlog out. So I asked him to walk me through the last clip he'd filmed that never got posted. He thought about it and said there isn't one. Everything he films goes up. Idea to published takes him about 20–30 minutes, and he edits it without friction. The real issue sits earlier. Nothing tells him what to film or when, and all his raw material lives unsorted in his camera roll. He isn't stuck on publishing. He's stuck on starting. The second surprise was bigger. In discovery he'd said he wanted to spend less time chasing retailers, so I'd put retail outreach under "not recommending." Then we talked about what would actually let him go full-time, and almost all of that math ran through winning new stores. The thing I'd written off moved to the top of my list. He's already half-built his outreach with AI, but he still sends every email by hand, and he dropped his daily target the day after he set it. Third: the business can't see its own split between online and wholesale. Wholesale invoices live in a different system from the shop, so the split is his best guess. He works with business processes in his day job. His own words about his own company: "we run on paper." I don't see that as a criticism. I think it's the most common pattern there is, because expertise doesn't carry over into your own business when you have no time left for it. The finding I was proudest of, a technical email problem I'd verified before the call, he parked. Not a priority right now. That's fair. Pushing it would have been about me, not him. I also told him I'm still in school. Nothing changed.

https://www.youtube.com/watch?v=EBx-XDzr3Lg

Keeping the cost per call down for high-volume agents is a constant fight. For a restaurant booking bot or an after-hours dispatcher, the LLM bill can easily make the whole project unprofitable The latest model drops from OpenAI and Anthropic look like an all-out price war. Simon Willison's analysis points out that GPT-6 Luna is priced at just $0.10/M input and $0.50/M output, half the price of its predecessor. This could make certain high-volume, lower-margin agents viable for the first time. The contrast is stark: Anthropic's new Claude Opus 5.5 is priced way up at $4/M input and $20/M output, suggesting a split towards commodity vs. premium models But a price drop doesn't automatically mean you should switch a client's production agent. The fundamentals don't change. You still need to run your evals to check for regressions in task completion and tool use. You still need to benchmark latency, because a cheaper slower agent is often worse than a more expensive faster one. The price is public, but the performance isn't. We have no independent benchmarks for how these new models handle complex agentic tasks. Is GPT-6 Luna as capable as GPT-5.6, just cheaper? Or are we trading performance for cost? Until we see detailed evals, it's just a number on a pricing page.