The Three Ages of Applications Software
Historical context is needed to explain how much changes now with AI
The application software market of the last 30+ years—including what an independent software company is, its core value proposition, the intrinsic value of the code it builds and why it’s better than others for a customer's needs, and how customers buy, deploy, use, and value that software—is forever changed—starting now.
If I were to hazard a guess, using our definition today of what is an Application Software Company (i.e., application software that employees use to be more “productive”), I predict with more than 50% odds that in just a few years*:
There will be 75% less employee-facing application software companies
There will be 50% less employee-facing application software categories
We may see a return to “winner take most” dynamics in the employee-facing categories that remain
Note—this prediction is different for software that is used to replace the work of an actual employee—i.e., digital workers—but more on that later.
As someone whose career has spanned from working at the fastest-growing on-premise application software company of its time (Siebel Systems) to the fastest-growing cloud applications software company of its time (salesforce.com) to now being an early-stage investor @ Bonfire Ventures in SaaS start-ups - this is definitely something.
However, first, some history of application software and the industry’s approach to the user interface is important to set the context for my conclusions.
THE ON-PREMISE SOFTWARE ERA
Hard to Deploy, Hard to Use, I Hunt & Peck
In 1997, I started work at Siebel Systems. There has been a lot written about the company (it had an interesting culture and was the company that led to the eventual expensing of stock options), but it was the first dominant CRM provider in the market. It was so dominant that it took only seven years to surpass $2B in revenue - on just $54m of venture capital raised. Regarding its user interface, it essentially was a graphical UI representing a logical collection of the underlying schema in the database. Think-list views, form views, action buttons, etc. Here is a screen-shot of its interface:
This UX was no different than many other UX’s of the time. As a user, my actions were more or less CRUD actions (create, read, update, or delete) around an individual record. In regards to what a user wanted to do - i.e., I want to go update the history of some type of interaction, I want to find an answer to this question, I would like to see these “x” quickly and do something with them - they had to translate in their head what they wanted to do and then “hunt and peck” across an application to go make that happen.
One could argue that all of these software applications were not only not helpful to users but were not meant to be—they were directed and sold primarily to management at a company who were excited to drive consistent processes across their company and easily track and understand important metrics across their company (i.e., in the case of CRM—what's my pipeline, what’s my forecast, which reps are underperforming vs. not).
Humans can only comprehend so much and navigate across so many different tabs and views in an application. Like many initial products, Siebel's product was more straightforward because it only had so many fields and tabs. Inevitably, they sold to larger organizations that asked for more capability, which Siebel built into the product to drive more new business or add-on sales. This translated into even more tabs, sub-tabs, list views, fields, etc. - thus rendering “Siebel hard to use”. Siebel was like many companies in that era - they would stay focused for the most part on a given market focused on a core set of users within a corporate function for the following reasons:
Similar to software companies today, you had to maximize your GTM efforts around a core set of personas, or else you would spread yourself too thin and never build a real market presence as a leading solution for that functional group of employees
From a product perspective, it was very expensive from an R&D perspective to try and build for many functional users at the same time and build both a functionally complete and market-competitive solution
The deployment model was incredibly expensive, especially for the customer. Customers would take months to implement the software, often spending 10x on consulting fees to deploy these products. Worse, when the ISVs shipped a new release of their products, a customer, if they wanted to install the latest software, would have to spend a lot of money on services to re-implement the customizations they had done with the software’s prior release. As such, software companies tended to stay in their lane and build real expertise around a given functional/ strategic issue they were solving to build credibility and relationships with the C-level owners of that function.
As a result, many different ISVs in this era focused on being the best for individual corporate functions/roles - i.e., Siebel for CRM, Peoplesoft for HR, Business Objects for Analytics, Hyperion for FP&A, etc. In addition, there were companies, primarily because they had been in the market for 10+ years, who emerged and attempted to provide full-stop solutions - like Oracle and SAP, which had started in financials and ERP and attempted to make headway into CRM categories but failed and ended up over time acquiring those solutions (Oracle bought both Siebel and Peoplesoft as an example).
The nature of the economics of the on-premise world was fierce. Customers would buy the software upfront in a lump sum payment and pay 10-20% annually after that for support and maintenance/upgrade releases. As an independent software company, if you did not sell enough new contracts in a given period, you could run out of cash much faster and be more prone to constant hiring/fire-stop-start decisions. These economics, along with the fact that System Integrators were super influential in these markets (as an SI could make $10 on services for every $1 of software sold) and the ability of the richer companies to throw more money at firms like Gartner to “earn” magic quadrant leadership, led to a “winner takes most” set of competitive market dynamics. A leading player in a category would tend to end up with >60% of a market, the 2nd and 3rd would take the remaining 30%, and companies that faded away quickly fought for the scraps.
THE CLOUD SOFTWARE ERA
Easy to Deploy / Harder to Use / Still Hunting & Pecking
In 2003, I joined salesforce.com. What a heady, exciting time to be alive - especially once I got past the lawsuit Siebel Systems served me for daring to go to a competitive upstart (that used to be a thing back in the day). Marc Benioff and a super hard-charging set of employees (many who joined from the old world - on-premise) set out to effect “the end of software”.
A very cute turn of words as salesforce.com was software, but unlike its predecessors, it was simple to deploy and upgrade, sold as a subscription, and boasted a meta-data customization layer, which meant that all the changes a customer made to its instance were always maintained during new releases of the software. Additionally, because salesforce.com’s initial sales product was super simple (4-5 tabs, no custom fields, no dashboards, etc), it was really so much easier to use than Siebel, and you could just sign up and trial the software online without hiring Andersen Consulting to get it working - genuinely revolutionary stuff.
It was a marvel - especially for the thousands of sales administrators whose jobs were horrible in the on-premise world as they were on point in handling all the downsides of the prior deployment model. These people now had a real ability to effect change and impact their organization - they could customize pages, create new fields, author new workflows, and, gasp, create their own dashboards.
It is essential to understand that before the cloud, this was IT’s job. In fact, I recall an exec sales meeting at Dell where I sat between IT and the Business functions as there was strong resistance to giving business users the ability to change the application to meet their specific needs - it was like sitting at a dinner table between Jesus Christ and Charles Darwin. These same sales administrators became important at companies (ever try to hire a great head of Field or Rev Ops these days?) and were the real lifeblood around the palpable energy one felt when attending a Dreamforce.
But now, 21 years after my first day “in the cloud,” I can’t say the cloud was better for the end user. In fact, I can easily argue it got worse.
I state this because while the deployment/upgrade process for cloud software was 100x better than its predecessors, the user interfaces of these products did not change at all. Here is a screenshot of the salesforce.com sales cloud:
Look familiar? We see a collection of tabs, sub-tabs, list views, form views, etc. As such, the basic contract between the user of this software and the software vendor is the same - I need to go execute CRUD actions and translate “what I want to do” into an expedition of “hunting and pecking" across your application and within a page. However, in addition to this same old UX, the rise of the cloud led to the following “amazing” innovations for the end user:
Easier customization - yay, it was exponentially easier for my company to add to the database schema and add new objects, new tabs, new fields, etc, for me to navigate through. Administrators could do this quickly and in real-time without even notifying me of the changes. That is not awesome. Too many times when I met with clients at Salesforce.com, I asked them why they had added 23 fields to their opportunity page and what the fundamental changes were at the company they were looking to effect change with by buying the product. Go focus on that and the minimum set of fields for their reps to enter in that allows you to inspect what they inspect around that process change in their deal review, forecast calls, etc.
Easier integration - which means, oh great, now that it is much easier to integrate with other software applications (i.e., the data flows appropriately), then my buyer may decide to look at the relevant business processes for my role and determine that it would so much better if instead of hunting and pecking across one application, I should do that across many different applications. According to our friend ChatGPT, the average B2B sales professional regularly uses 10-15 software apps to do/manage their job. For a user, it brings both horizontal and vertical complexity - i.e., some of these applications fill niche gaps that their CRM doesn’t do for their sales process, and some of these applications are ones I am stacking on top of my CRM application because it might be easier to use/drive better actual engagement. Lovely. All I wanted to do as a rep was go sell some stuff and maximize my W2.
Unlimited Buyer Choices - For any given buyer today, there are 100 software solutions they could consider purchasing to solve their needs. On my on-premise days at Siebel, we competed with two to three other competitors. In the early days of Salesforce, there were a few, at best, cloud providers in any category. What changed? One - it is much easier today than 25 years ago to build software (at Salesforce.com, recall that we ran our own data centers back in the day as there was no AWS or Amazon), given how much of the software development stack has been “service-fied.” Two - the amount of Venture Capital available to these companies is 100X what it was for on-premise companies as the lure of, even in sane valuation days, the NPV of 7 years of subscription fees is too hard to pass up, I guess. Finally, cloud companies don’t die overnight as on-premise companies do. I.e., given the recurring nature of their revenue, they can zombie on for years if they manage their cash appropriately and remain hopeful for some type of exit.
Now, I know many might argue with me on this point around usability and shout - “what about product-led companies”? I won’t disagree that these PLG companies understood all of the issues I have outlined and attempted to address them - but I would argue that what PLG companies do is narrow the scope of the problem being solved for a user and tightly interweave the UX and workflow around a significantly curtailed more minor problem to be solved. Yes - it’s better than the traditional “multiple tabs / pick your own adventure” user journey, but it is more a clever hack than being truly helpful to the user and whose primary benefit is an economic one - i.e., offering a lower price point to reduce buyer acquisition friction which means you can’t afford to throw much sales and marketing expense at that “purchase.” That being said, those companies that did and do best offer a fantastic user experience and can serve a broad set of an individual functional employee’s needs - a small list - have created iconic companies - ala Canva for creatives, Figma for designers, etc.
This over-purchasing of software and over-deployment of multiple apps for any given employee has simply gotten out of control. I would argue that individuals are becoming slaves to how they master their applications instead of being great at their craft and mastering it. Marketers focus on attribution and not amazing breakthrough messaging or tactics. Salespeople can update all the records in their multiple tools but don’t know the proper meeting strategy for a given call with executives at a hot prospect, etc. I would say it is not their fault - it is, to some extent, as someone who has been in this industry for 27 years, kinda my fault as it is all of my peers. The basic fact is that:
Our software is not actually that helpful to individual end users.
Our software does not actually improve the productivity of end users.
Our end users rarely use our software and want to give us a high five and say, “damn, thanks. You saved me so much time and allowed me to be much better at my craft and deliver better individual/team/company results”.
Until now.
THE AI AGENT ERA
Easy to Deploy, Easy to Use, Software Helps Me - Finally!
The first time I saw chat GPT in action, I knew it was big thematically. It meant to me finally that software companies could no longer be lazy at building their product’s user interfaces. That, when a user logs into an application, the software should be able to:
makes it super easy for users to do whatever they want to do AND
knows what matters to them and recommends things for them to review and or take action on, AND
offers to do some of those actions on behalf of the user
With the rise in generative AI and all the billions of funding invested into these foundational models, richer application companies would no longer have data scientists building proprietary models that barely move the world forward with “recommended next best actions”. Nor would you have less rich companies (at least in their mind) think - oh yes, I will get to that AI stuff later way down the road when I can afford to. That is a ‘game over, dude” mindset.
The AI models are now becoming a commodity that any software developer can leverage, and the race is on. What is the finish line for this race?
It is finally the real opportunity for the software interface of the last 30+ years to move from a set of CRUD actions where all of the cognitive work is done by the software user to one in which the software is genuinely helpful in users executing and mastering their craft.
It is interesting how many different reactions I receive to this notion, including often visceral smirks - especially if I use the word co-pilots or assistants or agents. Perhaps because they think I am just some other VC lost in the hype train. Perhaps because their view of agents is of earlier customer support bots that sucked and were just poorly disguised paths for companies to cut down on inbound support calls. Or perhaps because, if you are in the software industry, accepting this new reality changes so many things that it is better to compartmentalize it away - because to comprehend entirely means you have to throw away so many of the norms and shift your mindset and go forward actions immediately.
But when I explain it in the following way, people’s eyes light up, and they get it:
Ignore software for a second. Let’s just think about people. Think of the people who use your software. That person is trying to do some task. Imagine if that person could hire a dedicated person to help them do their job. What would a world-class assistant like that be for that person? What would they do on behalf of that user to make that user so much more productive? What questions could they answer for their new boss? What work could they do that their boss assigned them to do? If they were really good, what work could they do without asking for their boss's permission because their bosses trusted them?
Ladies and gentlemen, that is the new standard for a winning software user interface. It may be more than that—the new standard/baseline for a winning application software company—one that is as helpful to a user or employee as a world-class assistant they wish they could hire to help them.
We have already started to see the first waves of these products in what I call the meeting scribes. Just like Gong or Kaia is for sales reps using salesforce.com, or Blueprint is for therapists, or my meeting scribe (offered for free within my CRM Affinity) that automatically records my meetings with founders, enables me to be fully present in those meetings, summarizes the meeting appropriately and suggest open actions for me to take next, and automatically updates my deal/portfolio company records in the system. The second you use one of these systems, you feel that “I am never going back to not doing it this way” and “I would never use a piece of software that didn’t allow me to do that.”
A CASE STUDY IN ACTION - APPFOLIO
As part of the research for this study, I contacted Will Moxley, SVP, Product at Appfolio, a public property management software company with over $600M in ARR.
Their platform is for property managers who manage apartment buildings and single-family homes to do their accounting, maintenance, leasing, etc - it's like CRM and ERP for property management. I thought of Will as he had worked as a product manager at Siebel Systems, had worked for me as a senior product leader at Salesforce.com, and was now working at a leading vertical SAAS company - ie, if anyone understood the journey from on-premise to on-demand and the potential that AI represented as much as I did, it would be Will.
More importantly, Appfolio represents a great example of a vertical-specific company - something that emerged with the cloud - i.e., there were few meaningful on-premise vertical software companies. We at Bonfire are large investors in Vertical SaaS companies as they can quickly expand into adjacent product categories for their customers once they make customers happy. Additionally, as they are very focused, you don’t waste years and capital pondering who your ICP is and what use cases you should prioritize. In the new AI age, I would be much longer on vertical software companies than horizontal ones because they solve problems for a broad set of problems for a particularly focused user set.
The discussions with Will were eye-opening as hell. Let’s get into it:
To start, let’s look at Appfolio’s property management platform.
This looks like a salesforce.com for property management - list views, form views, tabs, reports, universal search, etc. So, nothing particularly exciting for the end user other than, as it is with many verticals software applications, the Appfolio app is comprehensive, and I don’t have to use 7 to 8 different apps along with it to do my job in the pre - Age of AI era. But, boy, oh boy, Will is the first to point this out - there sure is lots of screen real estate to navigate through. Until now.
Appfolio just announced its Realm-X AI interface and capabilities. Is this some form of PR-generating AI-washing? No—it is the future of application software. Before I dig into more about Realm-X, it’s important to understand Will and the entire R&D leadership team’s journey to today.
Before generative AI, Appfolio, like many larger software companies, had been invested in AI for at least 5+ years in point solutions using traditional machine learning - i.e., an automated leasing bot customers put it on their website that answered questions about the apartments (i.e., do you allow pets, what's the minimum lease term, what's the rent, etc.) and if it couldn't handle something it would hand it off to human being. The other thing they did that many software companies have done is charge extra licensing fees for all of these additional AI components - but Will and the team killed that approach and believed that AI should be embedded throughout the whole platform so that everyone started to see & use it.
When chat-GPT first came out, Will’s team immediately realized this was huge - in their mind, “it was just like this giant leap forward from what was possible before - the kinds of things that you know would have taken you a whole team a long time to accomplish, chat GPT could do far better and faster.” For them (a $600M ARR market leader), they saw it as an existential threat - either they went all in on making generative AI work for their users, or a start-up could come along and challenge the company. If they did get it right, they also saw it as the way to leapfrog their number one competitor in the market easier and faster than conventional logic would imply.
They then tasked their team with the following mission - reimagine our user interface as an agentic one that leverages LLMs to allow users to simply do what they need to do - using their own words. With that reimagining in place:
They taught the LLMs their data model, which is a lot easier for a vertical application (ie, they are just telling the LLM you are a property manager) than a horizontal one.
They taught the LLMs what all of the data in their data schema means and its purpose, and then ran multiple scenarios through the agents with data to understand where it fell short and continued to teach it with more and examples - kinda like what someone would do with a brand new hire at a property management company.
They also became an API-first company overnight because, absent all of their endpoints exposed, the LLM couldn’t “talk to it”.
With the LLMs now completely understanding Appfolio’s target user, metadata, and APIs, it was on like Donkey Kong. The LLMs even knew how to respond to users' requests for what historically has been a multi-step workflow—i.e., a linked chain of multiple CRUD actions—wow.
Here is the AppFolio’s User Interface now - with Realm-X:
Note - the new interface on the right. A property manager wants to create a new guest card (or a new tenant prospect). Before Realm-X, a user would have had to:
Go to the people tab and enter the contact info for Frank Johnson
They likely had to enter or find the proper source field, which was likely a required field to submit the record.
Then, likely add in a child object the property Frank was interested in with hopefully some good inline embedded search
And hit enter
With Realm-X, a property manager is never going to do that again.
Imagine another use case that Will demoed to us - which is a normal part of a property manager’s job. They need to be aware of national weather service alerts to ensure the safety of facilities/tenants. Let’s say there is a tornado coming to western Tennessee. Before Realm-X, a user would have to go through “hunt and peck hell”:
Go to the query builder and pull up a list of properties in Tennessee. They will likely need to copy and paste the zip codes from the weather service where the storm was headed.
They would then hopefully be able to pull from those accounts the child object of primary points of contact at those properties.
Then, they would likely export those lists and re-import them into the messaging portion of the product.
They then have to craft a message for those contacts and send them one by one or in batches. When the POC’s first language was not English, they would have to translate and send a separate message to those individuals.
With Realm-X, here is now the interaction:
User: There is a tornado coming to Tennessee. Show me a list of all the properties in these Zip codes.
Realm-X: here’s a preview of those properties.
User: Asks to send an email to all the residents at each of these properties in their native language along the lines of “batten down the hatches or get to safety, etc.”
Realm-X: Here’s a preview of the note and asks if it should be sent.
User: Yes, please do.
Realm-X: Done
In theory, Appfolio PM’s could have prebuilt a specific workflow automation around this before generative AI. However, with generative AI, they do not need to do that and can offer infinite flexibility to support whichever action the user wants.
This is far better for Appfolio’s customers. The first feedback from their users is that it saves them over 1 day a week from doing menial tasks—or over 20% of their time back to do what they do best—be property managers. It is already changing the fundamental economics of their business for their customers —the old rule of property management is that one property manager can handle 100 units.
Appfolio, in Will’s mind, is just getting started on this journey - they are going to next build recommended actions and automation that doesn’t even require the user to prompt - i.e., “there is a hurricane hitting these regions, shall I notify the property managers which then evolves” to “shall I just do this every time without asking you.” There is now, in that situation, no user interface - all of the prior hunting and pecking by a user has now just become automated without them ever having to worry about that again. Some founders might say - “wait, that has nothing to do with AI and could have been done better with rules and automation”. My take is they are missing the point - what would have required some user or admin to think of multiple use cases and then write workflows and rules and then seek buy-in from all users to execute is no longer - what the end user asks for (what a crazy concept), AI delivers.
Appfolio’s goal is for one property manager to be able to manage significantly more units, achieving a substantial increase in productivity from age-old industry standards. Appfolio is also super excited in that they believe they will cut down onboarding time by more than half and user adoption up dramatically as:
No longer does a new user have to learn how to use its software - the user just does what they want to do. It’s the software’s job to learn how to do it for them.
Onboarding and training now is reduced to explaining to a user / customer why they are using this software and what it can do for them rather than how to use the software.
Finally, and almost as importantly, is what Will directs his sizable R&D teams to go do from here on out. Are they going to go and devote countless hours to building new “functionality” the old way - ie new screens, tabs, and views that you then do a bunch of usability testing on, announce with banners in the product this capability is ready, update all of their support docs and faqs, train all of their success team on the new capability, and then leverage product analytics and guides to determine why people are not using it and nudge them to do so. Uh - no. That takes forever and kinda sucks. They are going to focus on ensuring 1) they understand precisely what additional actions their users want to do, 2) they are going to extend their data schema, 3) they are going to train Realm-X to understand the schema, and then they are going to rock n roll and move faster in beating their number one competitor.
Hopefully, as you have been reading this, you have been cycling through some of your own “oh holy s#$t” moments and asking yourself what this all means. It’s a lot to process and think through. In my next piece, we will attempt to pay off our bold predictions in this piece’s introduction and explain what we believe this means for the application software market in general and those founders trying to make their big imprints within it.
This is the best description of the evolution of application software and where it is going that I have read. Nice job Brett.
Brett this is a super exciting example of what all software platforms need to be thinking about. I would love some guidance on how you build this stack as an application for a community / influencer model. I have meet you a few times over the years . Would you be willing to give me some guidance