Programming

AI Is Eating Low-Code: Why We Ditched Our Platform for In‑House Tools

AI Coding AgentsBuild vs BuyDeveloper ToolingTechnology EconomicsLow-Code Platforms
Added Jan 27
Link157
Article: PositiveCommunity: NeutralDivisive
AI Is Eating Low-Code: Why We Ditched Our Platform for In‑House Tools

AI-assisted and agentic development erodes the core benefits of low-code by making custom in-house tools faster and cheaper to build and maintain. Cloud Capital’s switch from Retool to AI-enabled, in-repo internal tooling improved speed, UX, and total cost of ownership, prompting a rapid full migration. While low-code vendors are rebranding around AI, the author believes the build-vs-buy calculus now often favors building.

Key Points

  • AI coding and agentic development invert the traditional ROI of low-code platforms by making in-house development faster, cheaper, and simpler.
  • Low-code requires significant developer enablement (data plumbing, auth, custom components) and adds platform overhead, versioning friction, and UX constraints.
  • A real-world case at Cloud Capital shows a rapid migration off Retool led to better velocity, maintainability, security, and user experience.
  • Despite bullish market forecasts, low-code vendors must pivot to AI-heavy offerings to stay competitive, and it’s unclear if that will be enough.
  • For many teams, the build-vs-buy decision now favors building with AI due to improved speed, lower total cost of ownership, and reduced organizational complexity.

Sentiment

The community was broadly skeptical of the article's bold claim that low-code is dead. While many acknowledged that AI tools are genuinely shifting the build-vs-buy calculus for developers, the dominant sentiment was that the article overstated its case by conflating 'writing code' with 'shipping and maintaining software.' Most commenters saw a future where low-code and AI converge rather than one replacing the other. Experienced developers repeatedly emphasized that code generation was never the bottleneck -- operations, maintenance, security, and organizational complexity are what make low-code platforms valuable. The article was seen as reflecting a real trend but drawing an excessively sweeping conclusion from one company's migration off Retool.

In Agreement

  • Direct experience confirms the shift: one developer reported building internal tools with Claude Code faster and with better results than Retool or n8n could provide.
  • A low-code platform co-founder shared that a customer had already canceled their subscription in favor of an AI-based platform, validating the trend in real time.
  • Another developer planned to replace Budibase CRUD apps with AI-written code because it was even easier, and the resulting apps use standard components rather than proprietary ones.
  • Low-code's biggest weakness -- hitting a wall when trying to customize something small -- is eliminated by AI, which offers the speed of low-code without the flexibility constraints.
  • Low-code DSLs are far less optimized for LLM interaction than general-purpose code, and billions are being invested in general-purpose code-first AI tooling.
  • A former AppSheet employee described low-code as 'the worst of both worlds: complex, locked-in, not scalable, expensive, with small ecosystems of support for self-learning,' noting AI tools already offer a magical experience to non-technical users.
  • One commenter argued that proprietary low-code platforms can no longer lock in customers when AI makes it trivial to replicate their functionality with standard libraries and frameworks.
  • A verticalized low-code startup founder described killing their low-code product after ChatGPT 3.5 launched, recognizing the writing was on the wall for generic low-code productivity apps.

Opposed

  • Low-code and agentic tools will merge rather than one replacing the other. Visual introspection and direct manipulation let non-technical users understand and refine what agents create, filling each other's gaps.
  • Code creation is the cheapest part of software development. More AI-generated code means more reading, maintaining, and adapting -- exactly where low-code's abstractions provide the most value.
  • Low-code DSLs constrain LLMs in useful ways, keeping them on the happy path. Agents working with simpler, smaller sets of primitives produce more reliable output than those working with unconstrained general-purpose code.
  • The article's premise that 'the cost of shipping code approaches zero' is wrong. Shipping requires knowing how to deploy, monitor, and fix production systems -- LLMs do not solve operations.
  • AI-generated internal tools will face the same maintenance nightmare as any unmaintained code. Check back in 18 months when half those tools have drifted and nobody can figure out what the code was supposed to do.
  • Low-code platforms handle operational burdens like library upgrades, framework deprecation, hosting, compliance, and RBAC. Writing code is only part of the equation; running it reliably is the harder problem.
  • Enterprise organizations need deployment, hosting, compliance, and maintenance guardrails that departmental AI-built tools cannot provide.
  • The DronaHQ founder reported a customer who built successfully with Cursor but saw everything break when scaling to version 2.0, wanting to return to low-code for stability.
  • Vibe-coded internal tools will cost far more than original subscriptions once you factor in security audits, privacy compliance, maintenance burden, and operational risk.