1. Pricing AI is way harder than pricing SaaS

Classic SaaS? Infrastructure costs per user are negligible. AI? Every user query costs real money. A lot of founders are feeling anxiety here - so if you do, don't feel alone. Krzysztof Szyszkiewicz (Valueships) is your guy for deeper strategy, but I loved this quick insight from Rahul Vohra (Superhuman):

"Start by using the most powerful model available. You’ll quickly notice that most costs come from just 5% of users. Just find a solution dedicated to those 5% of users - custom api key, custom deals etc"

2. Speed beats perfection

Luke O'Malley started the conversation sharing his experience from Semgrep

TLDR?

We need to move fast. Don’t aim for 99% accuracy before launch.

Pick a group of trusted users. Test with them early.

Tight feedback loops > flawless pipelines.

3. You need AI champions (not just data scientists)

This one came up a lot.

Hiring PhDs isn’t enough. AI transformation needs people inside your org who are genuinely excited about what’s possible—not just focused on what’s not.

These champions don’t even need to be technical.

I heard from so many founders that their tech teams are overly focused on limitations. You need people who’ll experiment, not just analyze. (btw. bards.ai can help you there :) )

4. Black boxes break trust - You need to build in the open

If your AI feels like magic, users won’t trust it.

Show your process. Show your sources. Let users feel like they’d make the same decision the AI made.

Nicole Bentz from S3 Ventures outlined that from her expirience at S3, blackboxing is one of the top reasons ai-first solutions can fail.