AI Fails to Speed Up Cross-Border Finance Launches and Legacy Hardware Support

Hiring taking longer than before is a good one...

Adding Linux support being as infrequent as ever (even tho it should be easier!) is a good one to mull about

At least Claude Desktop added it, but its biggest competitor Codex has indeed not

@GergelyOrosz Adding Linux support. I still cannot use Codex on Linux. Have to fallback to VSCode to even reset my limits for example. And no CLI does not count.

@GergelyOrosz Software was never the bottleneck.

For 1), EU regulations and compliance are. For 2), I believe there is not incentive for that.

An interesting one that haven't got faster at all is hiring. Interviews got AI assisted on both sides and hiring timelines got longer.

Then

1) Why is it not faster for non-US fintechs to launch in the US? Why is it slow to launch in India, and all other markets? It's not just EU

2) OK, another example: adding Linux support that came up. Not happening more frequently, even though it would increase customer base!

For (1) EU regulations slow it down to a crawl. For (2) why would companies spent resources on a "long tail" device support - the same logic would apply to legacy s/w. The other industries relying on s/w but involving real life services seem getting more efficient but not necessarily faster - medical services, legal, retail - it is possible that bottleneck for "speed" is somewhere else.

@GergelyOrosz AI is great but kinda.. everything? Like I'm not seeing features drop faster or higher quality across any of the main production software I use, eg Android, Google web apps, GitHub, Slack, etc.

@GergelyOrosz Well I guess these are down to things other than softwares or infrastructure. I guess things like relationships won’t be influenced by AI too much either.

@leaving_tech @GergelyOrosz for hiring: understanding and establishing fit is complex and multifaceted

You can fasten parts of the hiring process (top of funnel screening, comms automation), but not the most critical ones, requiring taste / judgment

AI for interviews (esp. high stakes) makes no sense

@GergelyOrosz Stripe is still unavailable in most parts of the world and nothing moves for years now

@GergelyOrosz you can make so much software but users should be willing to use it. training users for use that app is also tough, mostly in enterprise users even tech professional also frustrate to use new software like i am very used to use cloudwatch i hate to use datadog bcz learning curve

@GergelyOrosz banking integrations are a shit show because most of the bank APIs are incomplete and poorly documented. lots of manual onboarding to get developer access, sandboxes etc. then also auth flows. obvs the agreements need to exists first, so that is friction also.

@GergelyOrosz Agentic payments. Agents face onboarding walls and paywalls. Protocols are there (x402 protocol by @coinbase, MPP by @stripe), but suppliers don't see this opportunity and volumes are low. Btw, @cloudflare recently added content monetization via x402.

@jmarceno @GergelyOrosz If coding was solved there would be three native clients for each platform. Yet still Electron, and still no Linux.

@GergelyOrosz Just don’t know if there’s customer pull for (2)

@GergelyOrosz I think it's mostly due to the checks and balances, and the compliance part that make it a long super long to iterate.

An other thing that takes very long is: 1. B2B sales cycles - seems like AI has barely made any dent here (most likely due to the previous momentum).

@GergelyOrosz yeah resetting limits in VSCode is so dumb

@GergelyOrosz Number 2 is intentional to "encourage" people to upgrade.

@GergelyOrosz AI is just accelerating the enshitification of things. There’s only so far that they can make their own products almost useless. Sadly the AI is also getting shittier to use as they continue to “iterate”.

@GergelyOrosz 2 is probably because it makes no financial sense to do that. You make more money by "forcing" people to buy new hardware