Build vs. Buy in the world of AI

Vibe coding shifts tides

Windsurf has been a crazy rocket ship, we’ve been hiring like crazy and the capabilities we need to scale the company are evolving like crazy. When we create new business units and teams, one of the main needs are to get the right tools and software to support them, but then comes the age old question:

“Build vs. Buy?”

Having seen multiple startups get started from the ground up, buying tools to quickly get things rolling are table stakes for a fully functioning team. There are tools that nearly every company needs, and these include CRM, messaging, support, finance and billing and so on.

However, as things keep going, you start adding more tools, what about recording and transcribing all calls for coaching, what about CPQ (quoting) and data visibility? What about taking notes and sending outbound emails? There’s an ocean of tools to buy, and they get added at almost an unlimited pace.

Now let’s start quantifying the cost of these tools, with some, thankfully they do not scale with the team on a per user cost. As you scale from 10 to 100 to 1000, multiplying per user becomes a sneaky issue. There are tools that are flat fees and are most of the time no-brainers as a result, but the downside of these tools are that they are usually quite expensive.

Let’s try to get some hard numbers around these tools, which is what every company has to encounter at some point, even if they don’t scale with the company. For the sake of simplicity I’m going to round up and down to use round numbers (negotiations and discounts apply at larger scale too), but I’ll give a few examples:

There are obvious productivity tools and email, these can easily go to $30 a user, we won’t count these because of the heavy lift to build and get a company started.

There a few tools we spun up quickly to get up and running, which unfortunately scale per user:

• CRM tools like Salesforce could cost ~$100 a user a month, or $1200 a user

• We also use Gong to record meetings and help coach our reps, this is surprisingly expensive, also around ~$100 a user a month

• Next up, what about enriching tools, these can be actually $50 a month! That’s $600 a user

• Sequencing tools like Outreach are surprisingly expensive too, you guessed it, $100 a user a month as well

At this point, I’ve only talked about 4 tools, and it already adds up to $1200*3+$600=$4200

$4200 isn’t bad right? But for 100 users, that’s already $420k

Oh and this isn’t even all the tools, when you start adding things like LinkedIn Sales Nav, that’s another casual $120k, great now we’re at $580k, oh what about adding a CPQ tool that’s $120k? That brings this back to $700k.

And…that’s not even all the tools, if I keep typing this newsletter, you’ll see we are going to get close to $1M spend per 100 users, that’s quite ridiculous. What about using no-code tools like Retool? That’s another $120k. What about observability tools? That’s another $60k.

When you have ambitions to become a 1000 person org, that quickly looks like $10M spend.

So you know where I am going with this, it’s literally the title of this newsletter post, what are the tradeoffs between Build vs. Buy? What if we could build away some of the demands that the org needs?

For example, if we could build a custom web app to do some simple things, we could mitigate some of the Retool scaling issues. What if we built some very specific web dashboards for different teams? Turns out that’s not really hard with Windsurf. Well that decreases the number of dashboard licenses. Before we keep listing off easy wins, let’s throw some more numbers here to get a sense of the problem.

Basically you can sort of draft a very simple formula of SaaS tools spend like this:

Cost = Num Users*Scaling SaaS tools + Fixed cost SaaS tools + SaaS app sprawl (who knows how much this costs?)

Time to value = The time you need the SaaS app to be working to provide value

Maintenance = The amount of resources needed to keep it up and running or updated

You’ll want to maximize business value and make all these trade offs worth it, meaning:

Business Value > Cost + Time to value + Maintenance

For Buying SaaS tools, thankfully the Time to Value and Maintenance are relatively close to 0, your Value better be greater than your costs. Therefore with the examples above, with 1000 users, the equation looks something like:

$10M = 1000*Scaling + Fixed Apps + SaaS app sprawl

That means:

Business Value must be > $10M + Time to value (0) + Maintenance (0)

But the equation is a bit different with Build, where you can control the scaling costs:

X = 0*Scaling + Built Apps + 0*SaaS app sprawl, however you now add Time to value and Maintenance:

Can Value > X + Time to value + Maintenance? Building makes it hard because it requires Time and Maintenance. In fact traditionally this is where Sales people will try to scare people that Build vs. Buy on most of these tools aren’t worth the time and maintenance. After all, there are many horror stories of hiring entire teams just to build an app for the organization, or even hiring a team just to maintain applications are they are being spun up. With SaaS tools, a small team can even manage many applications across the entire organization if run right.

But here’s where the value of AI tools come in, with Windsurf and the whole notion of Vibe Coding, you can reduce the Time to build these tools and even the amount of Maintenance. I’ve seen tools at larger companies take more than a quarter to plan and build, but after spinning up a few apps and teams internally, we’ve been able to bring that number down to roughly 1-2 weeks. With maintenance, a single person can make changes and deploy them live multiple times in the same day. Now, that’s not to say that some of the features are much harder to build, but when we plan on building Internal Apps now, are targets are closer to go live within 1-2 weeks and iterate on an ongoing basis. This drastically increases the favor to Build. It brings down scaling costs and also brings down time and maintenance, there are almost no downsides here (though we’ll get to some in a bit).

What have we saved money on so far? We built that CPQ app and saved $120k, but the more important thing is, we have been customizing it to exactly what we need for our org. The turnaround time is typically same day when we come up with ideas and we are not tracking things by months. We’re also making it integrate better with the rest of the Sales process.

We’ve also built partner dashboards and tracking tools for the partner team, now we’ve saved another $150k and scale the org literally for free as we add more employees.

We’ve given access to tools to reps that can do research on accounts, that will save us $240 a rep, or $240k if we were to scale to 1000 reps. Although I still see a world where reps are still using Gemini, Perplexity and ChatGPT.

We’ve created tools to do outreach, for example an AI SDR app to scale similarly to an SDR team, but this will cost us hundreds of dollars a month, not tens of thousands of dollars as we go bigger.

We’ve given tools to modify and query important data, again, no need to give things like retool to all employees, that’ll save another $100k or more.

Now the equation quickly becomes the cost of resources to build apps, whether it’s full time employees, contractors, or taking time from existing employees, the ROI equation is quickly overcome by building only 2 applications of decent value per person. With our equation, Time is reduced to 1-2 weeks, that means by hiring 1 person, you can quickly get ROI in less than a month as long as you have a lot of productivity apps to build.

There are obviously many varying degrees of complexity in apps to build with varying levels of value, for example we built a simple slack bot to track bugs into notion, but you could technically do that with Google sheets, however, simple doesn’t mean cheap, it’s been much more valuable to customize a solution rather than try to overfit available tools as well. We’re also building apps that do not exist in the market yet, but are specifically tailored for gaps or optimizations we find in our sales process.

That begs the question, do you have enough apps to build? You have to quickly think if ROI can be achieved, instead of replacing apps you would have bought, you have to think of what capabilities your org actually needs as it scales. You have to think about what kind of tools can be given so that your team can increase velocity building product or reducing repetitive tasks so that people can do more with their time.

By adding more and more leverage to the organization, you are creating way more value than $10M, you are generating revenue and increasing the total employee output faster than the number of employees you can hire at. This also gives you leverage against competitors, you’re supercharging the team and building faster than they can.

However, you’re probably wondering how to prioritize which apps to build. There’s a few simple things that either reduce the cost of buying apps or increasing the leverage that the current organization has:

• Foundational capabilities, apps that provide the data each employee needs to know to do their job, and using these app to push that data directly to each person at the right time

• Really doubling down on whatever makes someone good at their job. If someone is really good at organizing conferences for example, can we build tools so we can double the amount of conferences we can do with that person? What about Sales reps seeking connections with the right folks, can we double the pace at how we are doing that with tools? These are multiplying effects as you hire more people.

• Reducing repetitive work, so people don’t have to waste time on things that aren’t the best at what they are good at

In order to hit the ROI from a dollar perspective, the priority should be:

• Apps that you traditionally would always buy that are currently expensive, the good thing about these are you have so many good examples to copy from but can still customize to your organization (such as a CPQ app)

• Apps that scale with the team, if you reduce the variable cost to $0 because you have your own app, then as your headcount grows this app won’t increase in cost

There’s also other obvious use cases with these AI App builders, one is prototyping and iterating ideas, another is executing very concrete tasks to get things done, like running a complex query based off of a CSV file so that I can populate Salesforce with the right data.

There’s also downsides democratizing building apps though:

• While maintenance is much lower, it is not 0, as we build apps rapidly, the iterations and new ideas on existing apps starts to pile up quickly

• Errors and bugs will start becoming more common as non-devs start creating more and more apps, there’s a phrase that devs will transition from developers to reviewers, I can see that happening as there is more and more at stake with these vibe coded apps

• Standardizing security is of paramount importance, providing guardrails and guides to never expose data to the public will be super important, especially as we make deployment features more easier on Windsurf

• Extreme app sprawl, we’ll even have people building apps for their own personal use, it’s only a matter of time where there are many apps doing the same thing, this is potentially a feature, not a bug, however depending on the value and risk of the app, it’s better to centralize where possible

And on the flip side, what apps should you buy?

• Some apps have such a large ecosystem or ease of integration that it’s simply not worth trying to build. People joke about building Salesforce for example, but the fact it plugs into every other tool out there (including ones we build) makes it hard to replace

• Infrastructure or apps that require a lot of infrastructure, these will end up costing you money anyway, so sometimes purchasing them offloads that cost to the app itself

• Apps with a data advantage, for example enrichment apps would be very hard to build yourself unless you had access to all that data

• Apps that have a unique technology that isn’t available through vibe coding, though, admittingly, this pool of apps is getting smaller day by day as more stuff gets open sourced

Right now any curious employee can probably build 50% of an app today with no problem, they see something on the market and vibe code something very quickly that appears to work, but when you drill down it sort of falls apart. This means that data collection and querying is fine, but complex logic and persisting state could be more difficult for these newcomers that don’t have the patience to keep learning how to vibe code.

With the right context though, many semi-technical folks can get 80% there and get assistance from more technical folks, but over the next year or so, I can see even the complete vibe coder persona get close to 80-90% as well. The AI apps and models are improving so rapidly that it will only be a matter of time before building apps will be as common as spinning up a PowerPoint presentation.