alt Hacker NewsThat's really simple - actually writing the software has never really been the hard part in most SaaS apps. So long as you're moderately disciplined and organised it's easy to build what most SaaS apps are e.g. a CRUD-app-with-a-clever-bit. The clever bit is the initial challenge that sets it apart from the rest, but encoding that in software has never really been that difficult.
Having the ideas necessary to know what to write is where practically all the value lies (caveat: there is value in doing the same as someone else but better, or cheaper.) AI can help with that, but only in so much as telling you the basics or filling in the blanks if you're really stuck. It can't tell you the 'clever bit' because that is by definition new and interesting and doesn't appear in the training data.
What this means is that at some point Anthropic will be able to prompt Opus to clone Jira and never pay an Atlassian bill again. Opus just needs to figure out what Jira is first. It's not there yet.
Replies
On top of that, it's one thing to write the code, whereas it's another to actually run that code with maximal reliability and minimal downtime. I'm sure LLMs can churn out Terraform all day long, but can they troubleshoot when something goes wrong (as is often the case)?
I would posit another large factor is "owning" the software comes with the long tail of edge cases, bugs, support, on-call, regulations, etc... that an established SaaS has learned and iterated on for many years with many customers.
For the vast majority of companies they would (and should) rather let the SaaS figure that out and focus on their actual company
They just added task management in CC, which is a start in that direction.
This is what people don't get, what's coming up and it'll hit them like a ton of bricks. Software development, after toy examples, was a scale limiting factor for the better part of software development if you had domain expertise. Now, we hear constantly that it doesn't matter since "muh experience" and architecture, choices, tradeoffs etc for which you need seniority to operate LLM efficiently (or at all). This is true, of course. What people don't seem to get that that's what's coming next. Your experience won't mean crap anymore and then the ride starts full blast.
AI companies already know what they need. they're paying for it. it would make a great case study for them to make a list of all external software they're using, list the features they use (or make the ai watch them for a week), and then prompt the AI to rewrite those in-house.
> What this means is that at some point Anthropic will be able to prompt Opus to clone Jira and never pay an Atlassian bill again. Opus just needs to figure out what Jira is first. It's not there yet.
Bang on, and Jira is the perfect example! Because Jira isn't a bag of features: Jira is a list of features and the way they fit together (well or poorly, depending on your opinion).
That's the second-order product design that it's going to take next-gen coding AI workflows to automate. Mostly because that bit comes from user discovery, political arguments, sales prioritization, product vision, etc. It's a horrendous "art" of multi-variable zero-sum optimization.
When products get it right (early Slack) then it's invisible because "of course they made it do the thing I want to do."
When products get it wrong (MS Teams, Adobe Acrobat, Jira, HR platforms) then it's obvious features weren't composed well.
Expect there's more than one {user discovery} -> {product specification} AI startup out there, working on it in a hierarchical fashion with current AI now.