The Fatal Flaw of Vertical Software: Why Even Agentic AI Can’t Save Flooring Systems (and Most Trades Software)
Vertical software systems, especially in specialized industries like flooring, inevitably decline or fail over time. This isn’t because deep domain expertise is flawed, but because underlying technology platforms, economic incentives, and human/business realities evolve faster than most vendors can adapt. Even modern builds leveraging agentic AI will struggle for many of the same structural reasons. Here’s my take, informed by decades building and observing RollMaster/TMS in the flooring space alongside broader industry patterns.
The Historical Arc: From Mainframes to Desktop/LAN to Cloud
Computing for business software has come full circle in key ways, but each era introduced new failure modes for vertical systems.
Mainframe Era (1950s–1970s/80s): Centralized, expensive iron like IBM systems handled massive batch processing for big enterprises. Vertical software here meant custom or highly tailored applications. They “failed” (or ossified) due to prohibitive costs, lack of flexibility, and vendor lock-in. Small-to-medium businesses (like most flooring dealers) were shut out.
Desktop/LAN Era (1980s–2000s): The PC revolution democratized computing. In flooring, systems like early RollMaster emerged—born from real domain pain points in roll-goods inventory, job costing, pricing complexity, and supplier integrations. Launched from a spare bedroom with hand-built servers and Cox DSL, these encoded industry-specific math brilliantly. But they failed long-term due to fragmentation, maintenance nightmares, data silos, upgrade hell (on-prem), and scalability limits. Hardware/OS dependencies and events like Y2K exposed brittleness. Many dealers limped along with outdated versions.
Cloud Era (2000s–present, accelerated by Amazon AWS): SaaS promised elasticity, automatic updates, and lower costs. Amazon commoditized infrastructure. Vertical SaaS boomed in theory, RFMS (now Cyncly), QFloors, etc., but many stagnate amid subscription fatigue, integration sprawl, data lock-in, and vendor focus on broad features over deep excellence. Horizontal tools nibble edges while verticals chase tech shifts.
The pattern: Each wave lowers barriers and raises expectations. Domain depth starts as a moat but becomes a liability as platforms shift.
Why Vertical Systems Fail, Even Strong Ones in Flooring
Vertical software captures hard-won knowledge: flooring’s roll-goods math, remnant tracking, installer scheduling, pricing volatility, job costing accuracy, and relationship-driven networks. Persistent failure modes include:
- High customer acquisition and support costs meet low lifetime value. Solution fatigue persists. Solution fatigue persists.
- Maintenance and Evolution Burden: Rules change (new materials like RevWood, tariffs, regs).
- Vendors juggle deep code with horizontal demands (UI/UX, mobile, analytics), accumulating technical debt. Acquisitions often bring migration pain.
- Acquisitions often bring migration pain.
- Talent and Focus Dilution: Engineers prefer scale over “boring” verticals. Domain experts age out. Domain experts age out.
- Integration and Ecosystem Pressures: Dealers juggle multiple tools. APIs help but add fragility. APIs help but add fragility.
- Agentic AI Won’t Save Most, It Amplifies the Challenges, Especially in Trades and Self-Built Products
- Vertical platforms hold proprietary data advantages for fine-tuning. But it doesn’t fix fundamentals, and introduces acute problems in **trades** (hands-on, fragmented SMB sectors like flooring, construction, plumbing) and **self-built products** (domain-expert solo or small-team creations like early RollMaster):
But it doesn’t fix fundamentals, and introduces acute problems in trades (hands-on, fragmented SMB sectors like flooring, construction, plumbing) and self-built products (domain-expert solo or small-team creations like early RollMaster):
Many dealers/operators still rely on-site variables (weather, material variability, physical surprises), and low digital maturity. Agentic AI promises to automate back-office tasks (inventory, scheduling, customer service) so tradespeople focus on craft—but fragmented, incomplete data trapped in silos, messy manual workflows, and resistance to “black box” decisions for high-stakes job costing or safety-critical scheduling undermine it. Physical nuances resist pure automation; agents hallucinate on edge cases without perfect, real-time vertical context. Regulatory/compliance burdens (e.g., safety, contracts) demand human oversight that agents struggle to earn trust for in conservative trades cultures. – **Self-Built Product Pitfalls**: Domain experts excel at initial encoding of lived pain points but often lack modern engineering scale, DevOps discipline, or resources for ongoing AI orchestration.
Self-built systems accumulate technical debt that’s brutal to refactor into reliable agentic architectures (multi-agent swarms add complexity, latency, debugging nightmares). Maintenance falls on shrinking teams; talent for AI fine-tuning, governance, and monitoring is scarce and expensive. Lower barriers today let more self-built experiments emerge, but most can’t sustain the data quality, governance, and integration needed for production agents—leading to pilot failures, sprawl, or abandonment. Broader agentic issues compound this: reliability failures (rogue actions, high error rates on complex tasks), high costs for models/tools/monitoring, dependence on external providers, and governance gaps (auditability, compliance in regulated trades).
In trades, where relationships and “lived experience” drive decisions, agents risk commoditizing the very domain depth that verticals once owned, why pay premiums when general agents, prompts, and horizontal tools suffice for many workflows? History repeats: Platforms commoditize advantages.
Agentic AI may shift logic from brittle code to prompts/data/models, favoring adaptable generalists or data owners over traditional (or self-built) vertical codebases.
The Opinionated Bottom Line
They fail eventually because industries evolve, platforms commoditize, economics favor scale/adaptability, and inertia resists upgrades. In flooring and trades, survivors blend deep expertise with simplicity, open integrations, and AI as enhancer, not savior, while avoiding self-built traps through disciplined evolution or partnerships. For digital automation and control, prioritize unbreakable data foundations (clean, real-time, owned vertical datasets) as the true moat, AI agents are only as good as their context.
Adopt hybrid human-in-the-loop designs where agents handle routine orchestration but escalate exceptions, preserving trust in trades environments. Build modular, API-first architectures on modern cloud platforms rather than monolithic self-built stacks; this eases integration, reduces tech debt, and allows incremental AI enhancements without full re-platforming. Invest early in governance, observability, and reliability testing—far more critical than flashy agent features. Domain experts should focus on curating high-quality training data and workflows, partnering with horizontal AI infrastructure where possible instead of reinventing it. Ship relentlessly with ruthless simplicity: automate what moves the needle for dealer profitability (e.g., job costing accuracy, remnant utilization), measure ROI obsessively, and never lose sight of the physical, relationship-driven realities of the trades. Those who master this disciplined digital control will thrive; the rest will watch their vertical advantages erode.
Pure vertical codebases, even AI-augmented and self-built with passion, risk becoming relics. Domain experts who ship fast, iterate relentlessly, prioritize profitability, and govern AI rigorously will endure. The rest become cautionary tales or acquisition fodder.
Back to Blog