As AI tools like Claude Code and Openclaw made headlines in early 2026, a cascading fear gripped software investors globally. The fear was simple: if AI can write code, then software moats are dead, and every incumbent is a sitting duck.
What has followed is a broad, indiscriminate sell-off of software companies. The market is in a "sell everything" phase — treating the entire software universe as a single, undifferentiated risk.
While rising uncertainty and an accelerating rate of change have understandably pushed discount rates higher, the breadth of the sell-off fails to distinguish between businesses that happen to sell software, and businesses whose competitive advantages happen to be delivered via software. The difference matters enormously.
Where there is fear, there is opportunity.
Software Is a Delivery Mechanism, Not a Moat
Being a revenue-generating software company does not, in itself, constitute a high-quality business. Software is a gateway to a capital-light economic model that can deliver sustained high returns on invested capital — but only where specific conditions are met. These include durable competitive advantages such as network effects, unique and hard-to-replicate data assets, or a proven enterprise-grade capability in implementation, customer support, and ongoing R&D to solve complex, industry-specific workflows.
The key determinant of success has never been the software model alone, but industry structure — specifically where markets are tightly held by a leader and competitors lack the full set of ingredients required to compete effectively. The emergence of near-infinite coding capacity via Claude Code and other LLM tools does not materially alter this dynamic for strong software companies, though it does accelerate the commoditisation of many SAAS workflow applications.
Put simply: the market is right to reprice the weakly differentiated. It is wrong to reprice the structurally advantaged.
What Infinite Code Cannot Solve
The current fear rests on the deceptively simple idea that software businesses are primarily constrained by development hours, and that removing this constraint via AI levels the playing field.
In reality, development hours have rarely been the binding constraint for meaningful competitive differentiation. Consider the layers of complexity a well-entrenched vertical software leader has built over years, often decades, that no volume of AI-generated code can shortcut.
Take regulatory integration. A freight forwarding platform operating across 170+ countries must maintain live, certified integrations with the customs authorities, trade compliance databases, and tariff systems of each jurisdiction. These are not static APIs — they are moving targets, subject to regulatory change, bilateral trade agreements, and local government IT modernisation cycles. Achieving and maintaining certified integrations requires years of relationship building, in-country expertise, and a track record of audit-grade reliability. An AI coding tool can write an API wrapper in seconds. It cannot negotiate a certification pathway with the Australian Border Force, nor anticipate the downstream impact of a reclassification in EU harmonised tariff codes.
Or consider healthcare software. Clinical decision support systems, electronic health records, and medical device software operate within a web of regulatory approvals — TGA in Australia, FDA clearance in the US, CE marking in Europe — each with distinct submission processes, post-market surveillance obligations, and interoperability standards that require years of development and validation to achieve. The code is the easy part. The regulatory standing and clinical trust are not.
Enterprise data is another layer entirely. The most competitively advantaged software businesses sit above decades of proprietary, customer-contributed data: transaction histories, workflow configurations, industry benchmarks. New entrants simply cannot replicate these elements regardless of engineering resources. This data, structured over time within production environments, feeds machine learning models that continuously improve the product in ways a cold-start competitor cannot match.
Then there is implementation complexity. Enterprise software serving regulated industries: financial services, logistics, mining, utilities and government. This IP is typically embedded within the customer's operational workflows, integrated with legacy systems, and supported by dedicated customer success teams with deep domain expertise. Switching costs are not merely contractual, they are operational. A new entrant must not only build competitive software, but replicate the trust, integration depth, and institutional knowledge accumulated through years of live deployment.
None of this is to suggest that AI coding tools are irrelevant. They will meaningfully reduce development costs and timeframes for all players, including incumbents. But for the market to price in the destruction of structural moats on the basis that code can now be written faster, is to misunderstand what makes a great software business great.
The Opportunity Ahead
We continue to believe the strongest software companies will become materially stronger.
The medium-term pathway is the deployment of agentic AI solutions that materially reduce customer labour costs — genuine economic profit creation — alongside the effective monetisation of that value.
Companies with proprietary data, deep customer integrations, and domain expertise are best positioned to build and deliver these solutions. Companies without those assets will find that faster code merely accelerates price competition.
In Part Two of this series, we examine WiseTech as a case study.
The article has been prepared by ECP Asset Management Pty Ltd (ECP). ECP is a funds management firm based in Sydney, Australia. For further information, visit www.ecpam.com. This material has been prepared for informational purposes only and is not intended to provide and should not be relied on for financial advice. ABN 26 158 827 582, AFSL 421704, CAR 44198.
