Yeah, there's part of me that thinks Airtable could've done something more interesting here, where instead of becoming yet another vibe-coding app, they could've instead tried to be something more akin to the expert on top of the spreadsheets. Part of the reason the whole "Salesforce could be someone with a spreadsheet" bit is an exaggeration is because the spreadsheets do need to be interrelated and have some validations and all that, and Airtable does a pretty good job of that? And if you wanted to make an "expert with some spreadsheets" in a way where those spreadsheets were easy to read and understand, airtable wouldn't be a bad place to start.
I mean, there is this new pattern of Linear, Notion with tables with filter, display, grouping and saved views.
It is not so hard to craft one well but it always cost a bit outside of the real business you’re doing.
If you want a great search in your app, you buy algolia. Now if you want to have great tables/views… you should have been able to buy Airtable.
I’m wondering why there is no great solutions proving this as of now.
Even on automation, the field looks still early. I only see workato offering an embedded solution for this use case, and it is not even properly positioned yet.
Yeah, on the other hand (and probably the counter argument to my own airtable point), I'm not sure how much you'd really need that sort of platform? To take the dating app example from the post, do they need some special backend thing to host the database of users and user preferences? Part of the point of this structure could be to make the backend of the app mostly be a database, and if that's the case, you can probably just use a database? I'm sure there's some simplifying you could do for that, but it doesn't seem like it'd be all that much.
Yes, OK. I was not thinking about the backend part.
If we go by this example, this dating app will probably have to display a list or table to its end users at some point. And, to make a proper list/table, it will require efforts in the craft to be efficient at what it should do, as you put it: to compulsively check it.
For the users, yeah, I think the agent would have to check it a lot. Which is kind of the broader idea to me - it can do a thing a person would normally do every once in a while, but it can do it over and over and over again.
- I wouldn't think it'd be industry specific, but more role specific? Any sort of job where you spend a lot of time coordinating between things seems like a possibility. A lot of project management is fairly similar to this, or staffing jobs, or things like managing expenses or other back office IT management type stuff.
- As a broader point, it also seems possible that jobs themselves get...reorganized a bit? Saying something like, "this AI will replace sales ops roles" feels sort of unimaginative, because it assumes that tools fit into the jobs that people do. But it seems very possible that these tools eventually get good at parts of lots of jobs (eg, they're good at the part of your job where you synthesize a bunch of information into some more digestible piece of information and share that second thing with more people). And if that happens, you could imagine the boundaries around jobs are getting drawn differently.
An easy (but crude) example of this second point is the coding stuff. Bots can do parts of the job an IC engineer does. So does that mean we get rid of ICs and only have managers? Or get designers because engineers can make prototypes as fast as designers can make mocks? The answer to me seems like neither; instead, people just reorganize their engineering teams where some of the lines between engineer and designer get blurry and redrawn.
I have a very optimistic view of this. I think SaaS 2.0 can create many new types of jobs in the form of services as software. There is an endless number of problems left on the table currently in companies because they are too expensive to tackle. With SaaS 2.0 the cost comes down and floodgates open!
I do think there's something to that point, that there are a lot of service-shaped things (eg, reread every email and look for every time someone mentioned X, etc) that we don't do because they're too expensive to do, but would be valuable to do if you could do them for free. And people could start to do a lot more of those things.
This reminds me of the common B2B startups strategy you may have written about before:
You start off as professional services firm masquerading as SaaS. You sell into customers with "outcome-based" prices and then deliver outcomes before you have actual software built - usually using Excel and offshore teams. Then you slowly try build tools to either automate or let the customer manage (the latter is harder). If the company is good enough at automating, eventually the margins become tech-level.
In my experience, the last step is where companies fail. Compiling a bunch of overfit pseudo-platforms with custom code into a real multi-tenant software product is very challenging and involves migrations, "reimplementations" etc. But maybe AI removes this problem? e.g. if each customer can just stay a bunch of lists and custom logic that AI can run as part of a "platform."
I guess I agree that there's a decent chunk of software that, behind some sort of wrapper interface (maybe one of these AI browsers), can just be CSV files and list of expected behaviors/logic.
Using AI only for CRM castrates the capabilities of LLMs, since the ontology is already so well-established with high "network density". I can't see anyone vibe code a working sales playbook very effectively but business processes overall, yes.
Great read 🙏
I keep thinking that investing in a good craft for tables + saved views can help 95% of the user experience of SaaS businesses
And after this comes automated reports
Yeah, there's part of me that thinks Airtable could've done something more interesting here, where instead of becoming yet another vibe-coding app, they could've instead tried to be something more akin to the expert on top of the spreadsheets. Part of the reason the whole "Salesforce could be someone with a spreadsheet" bit is an exaggeration is because the spreadsheets do need to be interrelated and have some validations and all that, and Airtable does a pretty good job of that? And if you wanted to make an "expert with some spreadsheets" in a way where those spreadsheets were easy to read and understand, airtable wouldn't be a bad place to start.
True.
I mean, there is this new pattern of Linear, Notion with tables with filter, display, grouping and saved views.
It is not so hard to craft one well but it always cost a bit outside of the real business you’re doing.
If you want a great search in your app, you buy algolia. Now if you want to have great tables/views… you should have been able to buy Airtable.
I’m wondering why there is no great solutions proving this as of now.
Even on automation, the field looks still early. I only see workato offering an embedded solution for this use case, and it is not even properly positioned yet.
Yeah, on the other hand (and probably the counter argument to my own airtable point), I'm not sure how much you'd really need that sort of platform? To take the dating app example from the post, do they need some special backend thing to host the database of users and user preferences? Part of the point of this structure could be to make the backend of the app mostly be a database, and if that's the case, you can probably just use a database? I'm sure there's some simplifying you could do for that, but it doesn't seem like it'd be all that much.
Yes, OK. I was not thinking about the backend part.
If we go by this example, this dating app will probably have to display a list or table to its end users at some point. And, to make a proper list/table, it will require efforts in the craft to be efficient at what it should do, as you put it: to compulsively check it.
For the users, yeah, I think the agent would have to check it a lot. Which is kind of the broader idea to me - it can do a thing a person would normally do every once in a while, but it can do it over and over and over again.
Enjoyed reading!
Which industries you think will be the first to adopt the SaaS 2.0 model?
Consulting, professional services, agencies → Services as Software
If I had to guess, I'd say two things:
- I wouldn't think it'd be industry specific, but more role specific? Any sort of job where you spend a lot of time coordinating between things seems like a possibility. A lot of project management is fairly similar to this, or staffing jobs, or things like managing expenses or other back office IT management type stuff.
- As a broader point, it also seems possible that jobs themselves get...reorganized a bit? Saying something like, "this AI will replace sales ops roles" feels sort of unimaginative, because it assumes that tools fit into the jobs that people do. But it seems very possible that these tools eventually get good at parts of lots of jobs (eg, they're good at the part of your job where you synthesize a bunch of information into some more digestible piece of information and share that second thing with more people). And if that happens, you could imagine the boundaries around jobs are getting drawn differently.
An easy (but crude) example of this second point is the coding stuff. Bots can do parts of the job an IC engineer does. So does that mean we get rid of ICs and only have managers? Or get designers because engineers can make prototypes as fast as designers can make mocks? The answer to me seems like neither; instead, people just reorganize their engineering teams where some of the lines between engineer and designer get blurry and redrawn.
I have a very optimistic view of this. I think SaaS 2.0 can create many new types of jobs in the form of services as software. There is an endless number of problems left on the table currently in companies because they are too expensive to tackle. With SaaS 2.0 the cost comes down and floodgates open!
I do think there's something to that point, that there are a lot of service-shaped things (eg, reread every email and look for every time someone mentioned X, etc) that we don't do because they're too expensive to do, but would be valuable to do if you could do them for free. And people could start to do a lot more of those things.
Looking forward to read more of your writings on this. It’s an important topic especially the job-creation aspect: https://x.com/sama/status/1945541270438646270
Great read as always.
This reminds me of the common B2B startups strategy you may have written about before:
You start off as professional services firm masquerading as SaaS. You sell into customers with "outcome-based" prices and then deliver outcomes before you have actual software built - usually using Excel and offshore teams. Then you slowly try build tools to either automate or let the customer manage (the latter is harder). If the company is good enough at automating, eventually the margins become tech-level.
In my experience, the last step is where companies fail. Compiling a bunch of overfit pseudo-platforms with custom code into a real multi-tenant software product is very challenging and involves migrations, "reimplementations" etc. But maybe AI removes this problem? e.g. if each customer can just stay a bunch of lists and custom logic that AI can run as part of a "platform."
I guess I agree that there's a decent chunk of software that, behind some sort of wrapper interface (maybe one of these AI browsers), can just be CSV files and list of expected behaviors/logic.
Using AI only for CRM castrates the capabilities of LLMs, since the ontology is already so well-established with high "network density". I can't see anyone vibe code a working sales playbook very effectively but business processes overall, yes.
It does remember the Magic of Oz MVP style, but now made with AI agents.