I was thinking about how stressful modern interviews are right now. Whether you are an engineer, a PM, or in Sales, you are expected to have the perfect, charismatic answer to every behavioral and technical question instantly on a live Zoom call. If you freeze up for 5 seconds, you lose the job.
With local AI models getting faster and faster, we are quickly entering an era where candidates can (and will) use real-time AI teleprompters to feed them perfect answers on screen while they talk.
If candidates have access to tools that give them perfect charisma and technical answers in real-time, how are companies going to adapt? Will the traditional remote Zoom interview completely die out and be replaced by in-person whiteboarding again? Curious to hear what founders and recruiters here think about this shift.
Team size community engagement is the strongest predictor B2B categories (API, Payments, Fintech) convert at 3 the baseline Rank #1 on launch day 2.2 more likely to raise Series A
I finally set up proper conversion tracking this week. I had put it off for a fortnight because it felt like plumbing rather than progress. It took about an hour.The first thing it told me was that one of my funnel events has never fired. Not rarely. Never, not once in 28 days.People run a scan. People hit the gate. People hand over their email address to unlock the report. And then nobody clicks through to the app. The step I had assumed was the easy one, the one after they have already given me something, is where it stops dead.What makes it worth posting rather than just quietly fixing: I had spent the previous two weeks working on the top of the funnel. More traffic, better ad copy, keyword research. All of it aimed at a stage that was already working, because I had no way to see which stage was not.The part I keep turning over is that the absence of the event was the finding. Not a low number, a zero. A low number I would have argued with. A zero I could not.So, two questions. What did you find the first time you actually instrumented something? And had you been working on the right part of the product before you looked?
Been thinking about this a lot lately. The more agents get used for real work instead of demos, the more I keep running into (and hearing about) the same issue: an agent completes a step, reports success, and the output is actually wrong or incomplete. You don't find out until later, and by then it's harder to trace back where things broke.
Feels like this is going to become a bigger problem as more B2B tools ship agents into actual workflows instead of just chat interfaces.
Curious how others here are handling it:
Do you manually spot-check agent output, or trust it by default?
Has anyone built their own verification step for this?
Or is a certain error rate just the accepted cost of automation right now?
we build agents for a living, so a while back we pointed one at the least glamorous thing we had, which is cold outreach to store owners.
the useful part worked. it could find owner names by reading the replies people leave underneath their own google reviews, which is a far better signal than anything you can buy in a list.
Here's a frustration I'm running into: it's great that AI remembers context and follows the thread when you're deep in a topic. But when I want to pivot to something completely different, it keeps circling back almost too eager to stay "relevant" to what we discussed earlier. This makes the new conversation feel forced and unhelpful because it won't let go of the old context. I've tried deleting the chat and turning off data collection, but neither worked.
Anyone else stuck in this loop? Did you find a fix?
Estoy desarrollando un SaaS para complejos deportivos, enfocado inicialmente en reservas y gesti n operativa.
Estoy intentando validar el problema antes de seguir construyendo.
Qu m todo utilizar an para comprobar que el problema realmente vale la pena resolver y que los clientes estar an dispuestos a cambiar su forma actual de trabajar?
I wrote 40 posts about building k dlo before I wrote a single email to a stranger. And I know exactly why in that order. Posting feels like go to market. Likes, comments, followers, a nice little chart. But nobody ever said no to me, because I never actually asked anyone. I was publishing to people who already showed up. Selling is writing to one specific person who didn't ask for you. Building in public doesn't train that. Not even slightly. Which is why you can have 8,000 followers and three paying customers, and it isn't bad luck.
In 2025, being a marketer feels like being a full-stack SaaS team, minus the team. You re writing emails in the morning, editing video demos by lunch, tweaking automation flows in the afternoon, and debugging a landing page at midnight because growth doesn t sleep. And apparently, neither should we.
We told AI to make our jobs easier.
AI replied: Cool now you do everything. Alone. But faster.