TK Keanini, CTO, DNSFilter.
getty
As a CTO with over 35 years of industry experience, I’m constantly hunting for growth while strengthening my company's "technical moat."
A technical moat is a structural advantage that makes it hard for competitors to replicate your capability. With a strong one, they can’t reproduce your core product without significant time, capital or data.
Think of Apple. Their moat isn't just a phone; it’s ecosystem integration. Costly-to-replicate hardware-to-software verticality makes “leaving” the ecosystem a high-friction event.
For decades, moats were evaluated by replication time. If a smart competitor needed two years and $20 million to catch up, you had a strong moat.
The strategy for evaluating a moat is drastically different in the generative AI age. If you’re still using 2016 metrics to measure your 2026 defensibility, you likely aren't standing behind a moat but in a puddle.
Five years ago, a complex software architecture was a defensible asset. Today, GenAI has acted as a universal solvent, dissolving the barriers of "hard-to-write" code.
AI has compressed the time-to-market, but it has also shortened the half-life of a feature’s defensibility. In early 2024, an "AI-powered" feature felt like magic. By 2025, it became a baseline expectation.
The AI-capability frontier is moving fast. Stanford’s "2025 AI Index Report" found that on the SWE-bench—a benchmark for evaluating large language models on real-world software issues collected from GitHub—top-model performance increased from 4.4% to 71.7% between 2023 and 2024.
Meanwhile, LLM compute prices are falling drastically, with Epoch AI data showing them falling nine to 900 times annually. Specifically, GPT-3.5 costs dropped from $20 to $0.07 per million tokens in 18 months. At this velocity, AI system re-authorship is often cheaper than human legacy maintenance.
The economics of software creation have, therefore, shifted. AI doesn’t replace engineers, but it compresses what one engineer can accomplish.
GitHub and Microsoft researchers found that developers using an AI programming tool completed a coding task 55.8% faster than the control group. However, a study by METR last year showed experienced open-source developers working in familiar codebases were 19% slower with AI tools despite believing they would be 24% faster.
The implication is structural. The economic moat is no longer about headcount or difficulty. The constraint has shifted to what AI can’t replicate: proprietary data, embedded workflows and infrastructure that compounds over time.
In February, for example, a Cloudflare engineer used Claude Code to rebuild the open-source Next.js framework as a Cloudflare-native runtime (vinext) in roughly a week. A single engineer, paired with an AI coding agent, compressed work that would previously have required an entire team quarters of work.
These developments are part of the reason that industry leaders, including Node.js creator Ryan Dahl, believe that manual coding authorship is ending.
While software code is being commoditized, the value has shifted to assets that AI cannot "hallucinate" or generate out of thin air. These are the "survivor" moats:
Data is a liability if undifferentiated and a moat only if it’s unique, accretive and embedded in workflow. Think Tesla fleet data or Bloomberg pricing. For instance, my company has defined its moat around real-time DNS and identity signals, as they can’t be scraped or simulated, making them increasingly hard to replicate.
This is being so deeply tied into a company’s compliance, legal or operational "plumbing" that removing you is a nightmare.
The measuring stick here is asking yourself if you’re delivering a platform experience to that end-user's workflow, regardless of whether they’re pivoting across your products or in and out of your product.
Models are replicable; concrete, fiber and time aren’t. Durable advantage comes from capital intensity that deters fast followers, proximity to users and jurisdictions and hard-to-replicate regulatory constraints.
Hyperscaler spending, edge deployments for low-latency inference and tightly controlled environments all illustrate how physical presence and governance, not just compute, define defensibility.
GenAI is stochastic, which means its outputs are probabilistic, not fixed. The same prompt can yield two different plausible answers, whereas deterministic code always returns the same output for the same input.
Engineering leaders must decide where variability is acceptable and where it isn’t. That requires a systems-thinking mindset: Assume errors, bound their impact and keep humans in the loop.
The real advantage isn’t access to models but the operational experience deploying GenAI in production with safeguards, feedback loops and clear failure domains.
Your strategy must shift from building software to orchestrating advantage. Here are three things every CTO should consider right now:
Data gravity describes how large, high-quality datasets anchor applications and raise switching costs. Feature flash (i.e., chatbots, agents, summaries) is easy to demo and copy.
In GenAI, early annual recurring revenue (ARR) may come quickly, but differentiation erodes. Durable advantage comes from unique, compounding data generated through real workflows—signals competitors can’t scrape or recreate.
When your product team proposes a new "AI-driven" capability, don't just ask when it will ship. Ask: "If a competitor sees this on Monday, how many weeks of prompting and agentic workflows would it take them to have a 'good enough' version?" If the answer is under six months, it’s a feature, not a moat.
AI models’ half-lives are now shorter than the enterprise sales cycle. Therefore, your technical moat must be your architecture: the ability to swap models and components without collapsing your customer’s trust or your system’s integrity.
AI forces the question: Where in your architecture can you afford stochastic outcomes, and where can’t you? If done well, you’ll end up with a traditional set of deterministic processes interwoven with the new GenAI processes.
GenAI hasn’t eliminated the technical moat; it has simply exposed the weak ones. Your code can be commoditized overnight, but your data flywheels and global infrastructure represent the durable, compounding advantage that will define the winners of the AI era.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。