Joel Rybin

Oops.. i did it again.

[Joel Rybin @joelrybinn]

@AnthropicAI seems to be moving in the right direction and learning from its mistakes. OpenAI feels like it keeps moving backward. More and more of its decisions seem driven by profit instead of what users actually need.

I think @elonmusk had a point when he said OpenAI isn’t really open anymore. Its subscription plans are getting harder to justify, especially Pro. The 20x plan has been cut back so much that, to me, it feels more like 5x now.

[Joel Rybin @joelrybinn]

OpenAI’s pricing is getting out of hand for me. I can see how a $600 monthly plan with faster inference through Cerebras might be worth it for businesses, but that’s a lot to ask of everyday users. In some countries, that’s a full month’s rent.

I canceled my Pro subscription and plan to switch to Claude Max next month. It feels like better value for what I need. I don’t need the fastest model. I just want enough usage and consistently good output for my money.

[Joel Rybin @joelrybinn]

@haider1 Pretty sad... back to chinese providers or open source models.

[Joel Rybin @joelrybinn]

Right? I mean with their current models, not worth it. Overkill what they are doing.

[Joel Rybin @joelrybinn]

I’ve been an OpenAI Pro subscriber for a long time, but today I decided to cancel. Since people will probably ask why, here’s my honest answer.

OpenAI doesn’t feel like the company it used to be. Instead of moving forward, it feels like it’s taking steps backward. We keep getting “better” models, but in my experience, the quality hasn’t matched the promises.

Are the models more efficient and cheaper to run? Maybe. But as a customer, I care about the quality of the answers and what I’m getting for my money. Other AI companies offer comparable, and sometimes better, results at similar prices. Efficiency alone isn’t a reason for me to stay.

Now there’s talk of a $500 plan, and my question is: what exactly would justify that price? If a model comparable to Opus 5.5 ends up effectively requiring that tier because it burns through the $200 plan’s usage limits too quickly, that doesn’t feel like progress. It makes me wonder whether OpenAI has enough compute to deliver stronger models at the price its existing subscribers already pay.

I also don’t understand the reasoning behind keeping so many models available. How many do subscribers actually use regularly? Two or three? Would retiring models with very little usage help free up capacity for the ones people rely on? I know the infrastructure may be more complicated than that, but the overall lineup could still be simpler and more focused.

At this point, I’m seriously considering Anthropic’s 20x plan. For the way I use AI, it looks like better value.

I stayed with OpenAI because I believed in the quality of its products and the direction it was heading. Right now, I don’t feel the same confidence, and I can’t justify continuing to pay for it out of loyalty.

[Joel Rybin @joelrybinn]

@theo They will fight back with a 500$ plan.

[Joel Rybin @joelrybinn]

@MeanPanduh @edsheeran Free your own Country first, like all European who is saying "Free Palestine" before you start freeing other Countries.

[Joel Rybin @joelrybinn]

Peak-season prep worth doing this month: pull last year's top 10 ticket types and write the reply for each, while it's quiet. Come December, your team sends those same answers on repeat, under pressure. An afternoon now beats firefighting in November.

[Joel Rybin @joelrybinn]

@helpscout SMS inherits phone expectations, not email ones. A three-hour reply that looks fine in an email thread reads as broken in a text. Worth picking the response-time promise per channel before switching it on.

[Joel Rybin @joelrybinn]

@ExitEcom @CEO_Vlad It also leaves a trace in diligence: refunding chargebacks reads as triage, not a fix. What buyers actually value is repeat tickets trending down quarter over quarter, since that shows support runs on a system, not the founder's inbox.

[Joel Rybin @joelrybinn]

@liwei_w1 @MALOZY64 Most disputes die before the panel stage: the first-line rep wants to refund and can't. Give them authority up to a dollar amount, require a reason code, audit the outliers weekly. That fixes the common case; panels are for the rare one.

[Joel Rybin @joelrybinn]

@annieqyang One addition to #1: for support tools, the sticky part isn't the connector, it's the escalation rules and language coverage wired into it. Models swap in a weekend; a working playbook and the people behind it take quarters to rebuild.

[Joel Rybin @joelrybinn]

@rtehrani The open rates hide an expectation shift: people who get a text assume minutes, not hours, because every other text they send works that way. SMS pays off only when it's staffed to that promise. Otherwise it's one more channel where the customer waits.

[Joel Rybin @joelrybinn]

Signal 1 is the practical one: define what escalates to a human before the bot goes live (refund disputes, delivery exceptions, anyone on their second message), then automate what's left. Reverse the order and containment turns into the goal, which is how the horror stories get made.

[Joel Rybin @joelrybinn]

@Frontend_Prince A seventh fix worth testing: after-purchase trust. Return window, delivery expectations, and how fast support actually replies, all near the buy button. Wine shoppers pause on 'what if it's not right,' and the hero never answers it.

[Joel Rybin @joelrybinn]

@Chillestdotcom @JonTuckerUSA The overwhelm is real, but the fix order matters. The stores that stop drowning decide what humans keep first: disputes, damaged orders, anything already angry. Automate around that and the volume gets manageable.

[Joel Rybin @joelrybinn]

@SyloveO It keeps working after checkout, too. Customers forgive a late order; they rarely forgive asking twice and hearing nothing back. Repeat tickets per customer predicts the next purchase better than any survey.

[Joel Rybin @joelrybinn]

@3rdsonx Mostly two design choices: no route to a human, and context that resets with every message. Fix both and people stop complaining about the bot. Neither fix is expensive, they just don't register on a containment dashboard.

[Joel Rybin @joelrybinn]

Churn surveys say 'price.' The ticket history usually tells a quieter story: a question that never got a real answer. They stopped asking long before they stopped paying.

[Joel Rybin @joelrybinn]

@canipack21 @unhurriedself The version that survives growth: owners keep the high-stakes conversations and hand off the routine ones (order status, returns, address changes) early. Most stores do the reverse, automating exactly the conversations where tone decides whether the customer stays.