And on this, these are examples of tech companies filled with nerds who love novel new technology. Of course companies like Ramp and Stripe are spending huge amounts of time and money on taking this new technology to the extremes with “harnesses” and ”factories” because that’s what the nerds want to do. What the nerds want to do is not a sign of how the technology will be used in future, it’s a sign of what is most fun today.
So don't assume it's all nerds having fun. Now, is it what the future will look like for everyone? I don't think so, but that's because I expect AI is bringing us a more unified dev experience, with more developers than even Google has. This will make general programming harnesses very good, quite fast, solely because said harnesses can be differentiators in the AI race. So we are seeing some of the largest companies inthe world, with the largest research budgets, dedicating more money to the dev experience of their product than almost anyone else does.
You're talking about the best businesses. Arguably the article is talking about the mean of the "startup" tranche of businesses.
Google isn't a SaaS. That's (is or is tangential to) the kind of businesses this article is referring to. And they make decent money, they just aren't a unicorn that didn't die.
(This is temporary until the AI gets better judgement than humans, then capital will therever be the most powerful moat in a market full of dystopic, consequentialist, incredibly long-sightedly-greedy companies)
Long before A.I., tiny 10 person startups have been able to revolutionize industries.
Why wouldn’t capital simply lose its value even further, especially as capital globalizes further
The article argues (and I think I agree) that how you choose to incorporate agents into your work will be a differentiator and most great companies will have a unique take on it.
That's true for any product.
Who are they going to sell the product that this whole harness would produce? Any ideas?
1. The quality of Mattermost and Zulip was NOT exactly as polished as Slack.
2. Deployment of such stacks was a problem. A complicate operational overhead.
These days, that's no more the case. Agents can:
1. Make pixel perfect clone of Slack or Jira with SQLite or ejabberd behind.
2. Or they can deploy the Zulip or Mattermost or GitLab for you just give them an SSH key to the machine and see them bringing the stack to life.
So now and back then are not the same.
Mattermost isn't actually all that bad and its mental model is pretty close to Slack, at least compared to Zulip. I actually prefer Mattermost to Teams, and would view it in the same ballpark as Slack.
> Deployment of such stacks was a problem. A complicate operational overhead.
Not at all! Their stack is actually very reasonable and quite easy to setup: https://github.com/mattermost/docker/blob/main/docker-compos...
Nowhere near the nightmare that self-hosting Sentry is like: https://github.com/getsentry/self-hosted/blob/master/docker-...
I'd say that in Mattermost's case the group calling functionality was locked behind a subscription as well as some other stuff last I checked, which makes it dead on arrival for many. It was quite nice software though, even before being able to vibe code your own (though tackling videos sharing and RTMP will be anything but trivial).
> Make pixel perfect clone of Slack or Jira with SQLite or ejabberd behind.
There's A LOT of functionality and features in Jira, the only thing that saves claims like that is that you probably don't use 80% and can just build what your company needs. I more or less did that for the hell of it (MariaDB + Garage + Dropwizard/Java + Angular/Spartan) and there was still so much supervision an changes that were needed - after pulling the Jira DB scheme and throwing about 60% of it in the trash (didn't need the automation), there was still A LOT of stuff to do and even the plans eat up a whole bunch of the context of the coding tools you intend to use, no matter how many checks and scripts you write to automate guardrails for when the models simply don't recall everything that is relevant.
For people who want to self-host something basic, there's already the excellent https://kanboard.org/ (very very fast)
For those that want something similarly depressingly slow to Jira while still locking some stuff behind a subscription (at least in the versions I tried) there's the passable https://www.openproject.org/
If those don't fit and you don't see anything exactly like what you need, you might try to build your own. But yeah, not a weekend project.
I think you missed the part where Google was successful because it had the best engineers, and then went all Alphabet.
There is such a thing as too big too fail.