The age of operations
Historically, the codebase for your product was your most valuable startup asset. However, software products are now extremely easy to copy. Code is now cheap. Anyone can sign up for a free trial to your software, point an agent at it, and copy every user-facing feature.
To many, this looks like the death of technology as a moat: the only remaining moats are commercial, regulatory, or maybe patent driven.
This is certainly a change in competitive dynamics, and indeed, the relative value of commercial, network-driven, and regulatory moats is increased. But intellectual property is not dead, it’s just shifted elsewhere in the business.
The technology in your startup can be bucketed into the following groups:
- The customer-facing user experience. This is what the user can see and experience when they interact with your product. This is now incredibly easy to copy. AI has already commoditised it.
- Your application backend. Many application backends are simple to reverse-engineer from the user experience layer, and therefore easy to copy. But not all. Products with opaque backends, that abstract away innovative solutions to complex problems, are still difficult to replicate.
- Your operational toolkit. Historically, this has been incredible off-the-shelf at most startups. Tools like HubSpot, ZoomInfo, VSCode, Vercel, Claude Code. But today, with code being cheap, startups are increasingly building custom internal operations stacks. These can be impossible to see from the outside, and create a sustainable competitive edge.
So, startups should:
- Invest more in solving ambitious, complicated problems in their application backend, away from prying eyes.
- Invest much more in their operations stack, with custom software and intense automation. Importantly, you won’t get a competitive operational edge on your peers if you’re using off-the-shelf tools. Software must be personalised.
- Automate customer service.
- Make it possible for a CSM to own twice as many accounts.
- Accelerate feature delivery, per engineer, by five times.
- Practice increased secrecy. More software should blackbox in nature, and internal operations should be treated as the real trade secrets.
- Employee retention is more important than ever, because employees take operating secrets with them. Smaller, more loyal teams are now ideal.
Let’s run through some examples:
- Claude’s user interface is incredible easy to copy, and has been copied many times over. Claude has a legitimate backend moat (it’s LLMs). The team at Anthropic has an operational advantage by having access to models early.
- Cloudflare’s user interface has no moat at all. Their backend is powerfully defensible, given their global CDN network. Operationally, they should build custom software to accelerate their advantage.
- SendGrid’s user interface and APIs could be copied overnight, no moat at all there. But their backend abstracts significant complexity in managing deliverability. Operationally, they should build custom software to accelerate their advantage in managing deliverability and selling to new customers.
- Applied AI providers who currently require forward-deployed engineers could get a serious edge on their competitors through operational innovation that removes this need. If, for instance, Harvey continues to require forward-deployed engineers to go live, while Legora manages to automate this through an opaque software stack, Legora should expect better adoption.
Potential implications of this new reality:
- Only vertical SaaS makes sense now.
- It’s conceivable that there is little ROI in a business replacing a software product tailored to their specific problems and industry.
- But general purpose software (HubSpot, Notion, Slack) are different. Their customers have a strong incentive to replace these tools with something personalised for their needs.
- Software could end up a lot more niche, with shrinking TAMs for small players, and a lot more segmentation and market personalisation for the big players.
- Pain tolerance is a moat.
- Nice-to-have products are no longer relevant in B2B. You need to solve something that, even with the help of AI, is really annoying to solve.
- The only way Claude can save you from a DDoS attack is to sell you Cloudflare or something similar. You’re not going to vibe code yourself out of that problem on your own. The more exhausting it is to solve the problem, the better.