Brand As An Enduring Moat And The Post-MCP World
Branding in enterprise software, value accrual in application software
Software Synthesis analyses the evolution of software companies in the age of AI - from how they're built and scaled, to how they go to market and create enduring value. You can reach me at akash@earlybird.com.
As you scrolled through a timeline of Ghibli images, you wouldn’t know Google released the world’s leading reasoning model the same week.
OpenAI is the accidental consumer tech company with a brand moat that’s been repeatedly fortified since the ChatGPT moment.
A $300bn valuation at a c. 20x forward revenue multiple for the clear market leader actually looks cheap relative to the multiples other research labs are pricing at. The Softbank led $40bn financing would be the largest private financing ever and larger than almost any IPO in history.
Billions in losses and concerns around unit economics ought to work out in the long-run as the company continues to increase market share in the consumer AI market and leverage that to win in the enterprise (on the premise of the consumerisation of the enterprise) too.
It’s no surprise that the company that best personifies one of the few remaining moats in software is both a consumer and enterprise company.
Bill McDermott, ServiceNow CEO, recently came on Acquired2 and shared why he wants ServiceNow to be one of the top 10 brands globally and why he admires Apple:
My dream in the not too distant future is you already see us climbing the leaderboard toward the best brand, but I believe that ServiceNow has the potential to be a top 10 brand. I could say number one, but I have to say top 10 because in this world of consumer brands and the notoriety of consumer brands, most enterprise companies aren’t well-known.
Google’s a great brand. Amazon’s a great brand. Microsoft’s a great brand. Apple’s a great brand. All of them do have an enterprise side.
The reason I probably just came with Apple a little bit is just for my own bias because I have two in my pocket. Whether I’m doing business or I’m on my private time, I’m somehow associating with that brand. Then that brand can take you into the Apple+ experience.
Those very words could be used to describe the trust and brand conferred to ChatGPT by employees when an enterprise license is rolled out across teams.
The consistent tagline Bill uses to describe to ServiceNow is as the defining enterprise software company of the 21st century. There’s an inextricable link between the persona the company sells to (CEOs) and the artefacts of the brand.
To be a brand-led company is what it’s all about, because the brand is your identity. It’s your DNA. It’s who you are. It’s really what you’re trying to convey to a broad group of customers in a global economy in the simplest way possible.
Our dream tagline for the brand is the world works with ServiceNow. That was my way of tying it to the defining enterprise software company of the 21st century. But you can’t be that unless you’re the world’s company. And you have to make things work.
When Ford works with ServiceNow, the world works because Ford does so many things for the world. Always seeing the world through the customer’s eyes and what they’re trying to do, and being a brand-led company that has empathy for the customer is what we want it to be all about.
Why do I say the CEO is the preferred option? Only because they’re paid to have a broad vision of the whole enterprise. They’re paid to do what’s right for the enterprise, the people, the customers, and the shareholders, without any appendage to the past.
Michael Dempsey wrote about brand moats at the end of 2023, comparing brand moats of crypto protocols with emerging examples in AI:
Crypto has faced quite different dynamics to traditional OSS, with protocols facing consistent forks as the crypto industry tries to financialize public goods and open-source value creation. This leads to a rolling ball of money that is incentivized to rotate between projects and usage that is mercenary at best.
However, as we’ve seen over the past few years, this hardcore financial incentivization rarely pulls developers or users for a long period of time. This is because protocols that are longer-standing or truly novel have done a great job at generating brand moats in a fairly recursive and tight-knit space. Sometimes this is due to superior technology, but other times it is merely elegance of design, Lindy effect, community, and a variety of other intangible factors.
Model distillation comes closest to protocol forking in AI.
Although debate will continue about how OpenAI maintains pricing power in a world of abundant frontier models, Sam’s announcement of a soon-to-be-released open weights model suggests that they’re going to lean into their brand to become a 1 billion daily active user destination site.
For early stage, pre-PMF founders, brand-building is less a function of spending marketing dollars on brand awareness (though there is a place for that post-PMF), but rather a series of product design choices.
It’s important to note that brand moats are not built quickly, but instead are a result of many intentional steps by companies and the people that comprise them.
Think of Cursor and Granola - both went viral as a result of product design decisions that drove referral growth through the roof. The early days of ElevenLabs saw incredible growth among creators and the company soon developed a brand as the voice AI company.
Product matters, especially in the early days.
In this new paradigm, vibe-coding and lower barriers to producing software will simultaneously punish stale products and reward artisans that rethink consumer experiences with AI at their core.
Demand for the latter is insatiable at the moment and the teams that can best capture the zeitgeist will see stickier users and stronger retention than many think…
Application AI in the MCP World
In the six weeks since we discussed Anthropic’s MCP, the momentum has been deafening.
I installed a few MCP servers last weekend myself: Zapier, Puppeteer and Whatsapp.
There are a few directories of MCP servers but its clear that discovery remains a big bottleneck:
Dharmesh Shah and others have been equating MCP’s significance with that of ChatGPT itself, suggesting most of us are yet to have our ‘MCP moment’ where we realise the power of exposing any and all datasets to our preferred client (e.g. Claude, Cursor, soon ChatGPT).
The flywheel at work here is that as developers and consumers adoption grows, pressure increases for vendors to create official MCP servers. In the interim, the community is building servers that no enterprise is going to adopt because of security issues.
The consolidation around MCP in such a short span of time suggests it has won the right to be the integration layer of future AI workloads and that commensurate investment will be made to make it secure and enterprise-grade.
As we inch closer to that point, it’s worth reflecting on how this affects application software. Sam’s already said that owning the one billion DAU destination is more important than the SOTA model - well, if one billion DAUs orchestrate work inside of ChatGPT with MCP servers piped into all kinds of software (CRM, ERP, HCM, ITSM), what happens to switching costs and defensibility for these downstream companies?
Having barely scratched the surface of this myself (partially as a function of a lack of secure servers at the moment), it’s easy to contemplate a world where I complete all of my work inside my preferred OS (the MCP client) which takes multimodal input and frees me of ever having to log into any specific piece of software.
The residue of this is data. That’s the remaining asset. Hence all of the discourse on the value of proprietary data and new data generation as a way to differentiate.
Taking this scenario to its extreme, how else can application software vendors stop themselves from becoming fungible databases?
My working theory is that the drivers of brand moats in the early days (product design) create enough value for end users to still prefer to use that software to complete work rather than have it be abstracted away.
The bet is on software vendors at very least earning that right, and on the upside claiming the right to become the MCP client themselves - i.e. the de facto control centre used to complete work.
Additional Reads and Content
A Deep Dive Into MCP and the Future of AI Tooling by Yoko Li, a16z
Jevons Paradox: A personal perspective by
Roadmap: Data 3.0 in the Lakehouse Era by
Cybersecurity Investment Playbook by Gili Ranaan on Invest Like The Best
How ARM Became The World’s Default Chip Architecture (with ARM CEO Rene Haas) on Acquired
Jobs
Companies in my network are actively looking for talent:
An AI startup founded by repeat unicorn founders and researchers from Meta/Google building 3D foundation models is looking for a 3D Research Engineer, Research Scientist and ML Training / Inference Infrastructure Engineer (London or Munich).
A seed-stage vertical AI company building in accounting is looking for founding Product Engineers (London).
Reach out to me at akash@earlybird.com if you or someone you know is a fit!