Why? Getting operational experience with the product that you might then pay a lot for seems very important. Especially if you end up liking the product/service but not the pricing changes that might then happen, so doing some exploratory fact finding for a backup plan doesn't seem to be waste of time.

For example when we used Jira on-prem and it was snappy and we were happy ... and it was a rather important point of difference compared to the slow shitocumulus version.

Also, when people are using GitHub issues to ask questions the problem is usually a lack of clear documentation. (And if spending time to link FAQ answers to potential customers is a waste of time ... then maybe it's not surprising that Sourcegraph CEO is doing damage control on HN instead of focusing on focusing or whatever.)

> But it’s usually a huge waste of time.

Is it? I think at this point my company has probably saved millions of dollars by not paying for subscriptions, but hosting everything in-house. The price point of a lot of these services makes perfect sense when you are small, but paying 1M/year in subscription fees when you can host the same thing for 10k/year is just bonkers. I appreciate that someone has to pay for it for them to continue making the product, but there’s a point where it makes more sense for me to spend a year setting it up (and really only costs two weeks).

My experience was with things like openstack and kubernetes. The org decided to do “cloud” in house first with openstack and then kubernetes - and run critical services on them that had very strict performance SLA.

The amount of time needed to do the whole thing wasn’t worth it. Sure I enjoyed tinkering with the kernel and drivers and k8s. Also diving into known cgroups and namespaces worked etc. However, from a time to market/stability perspective the solution was nowhere comparable to what public cloud providers offer.

Yeah - the subscription costs more. My experience has been that when things get big and hiring gets tense in house solutions just add stress on the devs maintaining it. At least with public cloud services - it’s clearer - if the budget doesn’t exist don’t run it.

I will add that I don’t use sourcegraph nor am I connected with them in anyway. So I’m not batting for their go private strategy. Just commenting on this one point.

That math only works out nearly that cleanly if you avoid pricing out the engineer time for it.

If you’re paying $1M/year in fees, I would be shocked if you don’t have a whole team to support the open source version. Oncall, system upgrades, the usual stream of tickets about things not working right and people wanting to integrate, etc.

I do believe it can be cheaper to self-host, but I really doubt the difference in cost is 2 orders of magnitude. I’d be surprised if it was a single order of magnitude. I would wager it’s less than the sellers profit margins because of economies of scale; I would guess in the range of 10%-20%.

Many organizations will accept defacto 90% SLA if internal vs 99.99% If external

Well that obviously doesn’t apply to Sourcegraph because their self-host offering requires paying a subscription. You can’t use any form of Sourcegraph on private code, (at least not without all the important features being nobbled) without paying a subscription. So there’s no saving to be made from self-hosting sourcegraph

> So there’s no saving to be made from self-hosting sourcegraph

That may have been true in the time before LLMs, but I'd argue that any sourcecode exfiltration nowadays runs the very real risk of "oh, sure, we won't use your code for training our model ... wink, wink"

There's enough open source code available that the hassle of other code probably isn't worth it. LLMs aren't insightful enough to benefit from any differences you'd see between open and closed source code corpii.