what does durability even mean?
an essay on mastering the new game of durability
Last week, I wrote about how competition in AI has become more brutal than ever, and why the traditional advantages of speed and focus are getting commoditised.
Similarly this week, Ethan Ding also wrote a great article on "levered beta" that captures the current competitive climate in terms that a finance bro would understand. TLDR; companies and investors are essentially betting on foundation model improvements with different levels of leverage rather than building genuine technical advantages.
He also succinctly summarises the two options open to AI companies at the moment, with Option 2 being the better choice:
option 1: wait until the llm actually works, then scramble to build your chatgpt wrapper alongside everyone else who just realized the same thing.
option 2: start now, while the tech is still garbage. lie about how good it is. burn money on marketing. claim the territory while everyone else is still laughing at you.
The messaging is extreme, but also largely correct.
The Lovable vs. Replit battle is an interesting illustration of this.
Lovable just hit $100M ARR in eight months (slight caveat being they launched a few times with little traction, so have really been around for 18+ months). Replit took seven years to reach $100M ARR, finally reaching the figure last month.
Despite grinding away, building far superior backend infrastructure, and having an enormous head start on customer acquisition, Replit was really only able to exponentially grow and monetise from the same tailwind as Lovable, namely the launch of Claude Sonnet 3.5.
Whilst Lovable didn’t necessarily wait until the LLMs got better, (they had tried and failed prior to Sonnet 3.5) they did time their third launch to perfection meaning they were best placed to ride the tailwind, similar to other competitors like Bolt and Cursor.
On the other hand, Replit had the user base and a more sophisticated product, yet only saw a 10x increase in ARR ($10M to $100M) over the same time period. (As a side note, it’s wild to me that I just wrote that sentence with ‘only’ but such is that time we are living in).
From a pure business strategy perspective, Replit clearly missed an opportunity to capture more revenue given their head start. But when every company benefits equally from the same model improvements, timing and positioning become everything.
When foundation model improvements lift all boats equally, traditional competitive advantages dissolve. To build something durable in this environment, AI startups need two strategic capabilities to avoid commoditisation: Counter-Positioning and Process Power.
product design → business model → counter positioning
This is where we are seeing the most change, but also where there is less durability created than ever before.
Traditionally startups could play into the whole Counter Positioning narrative where they were building products and solutions that large incumbents couldn’t do. That’s kinda out of the window, with incumbents able to move quicker and reach feature parity much quicker regardless of core business model.
Startups have gotten too good at being memetic and following playbooks. The current memetic trend is to have taste. Everything from individual components, to the header to the typography needs to look ‘clean’.
My view is that this is completely missing the point.
To support my work and unlock the full article, become a premium member now. For just $5/month, literally less than a cup of coffee, you'll get the fundraising frameworks, distribution playbooks, and scaling systems that help founders build startups that turn into unicorns.
Keep reading with a 7-day free trial
Subscribe to Superfluid to keep reading this post and get 7 days of free access to the full post archives.