This document covers, at a high level, VMG Software’s Threats from LLMs/AI (2026/7 Landscape) and Prescribed Responses to Become the Most Admired & Respected Provider in Southern Africa.
It must be noted that this research was conducted using LLM. All LLM’s used (ChatGPT, Gemini and Claude), have promoted the further use of AI. This may be a biased view from their perspective.
VMG’s 500 customers, deep domain expertise in SA-specific dealership/workshop operations (accounting, VAT 264, 2nd Hand Goods Act, license disk scanning, Seriti/TransUnion/OEM integrations, etc.), support infrastructure, and local trust form a strong moat. LLMs lower barriers to entry and raise customer expectations but do not easily replicate reliability, accountability, integrations, or battle-tested local compliance.
The real risks are erosion of perceived value, faster competitors (local or new AI-leveraging entrants), commoditization of features, and pricing pressure—not overnight replacement. Below is a consolidated, prioritized list of threats synthesized from the Gemini, ChatGPT, and Claude analyses you provided, plus current market context. Each includes prescribed responses to convert threats into opportunities, with actionable steps for VMG product development and roll out.
- Threat: Lowered Barriers to New Entrants & Faster Feature Shipping by Competitors (Threat Level: High)
Small teams (2–3 devs + agentic AI tools like Cursor/Claude Code) can build functional MVPs in weeks, replicating core DMS/WMS features. Existing competitors (e.g., those integrating AI like Pinewood.AI, Autologica Sky with AVA) ship 5–10× faster. New “AI-first” point solutions could nibble at edges (e.g., workshop modules, marketing tools).
Responses / Opportunities:
- Accelerate your own velocity 5–10× using the workflows in the videos you shared (Axel Molist on supervisory role + Matt Pocock on grilling/PRD/vertical slicing/autonomous agents/sub-agents). Adopt agentic AI internally for feature development.
- Build a robust public API/platform so VMG becomes the “operating system” and source of truth. Let others (or AI tools) build on top while you own core data, accounting, stock, and compliance.
- Launch an open ecosystem or marketplace for SA-specific extensions (e.g., integrations with local insurers, fleet trackers).
Action for dev meeting: Audit current architecture for modularity/testability (per Pocock video). Implement sub-agent patterns and TDD for AI-generated code. Set quarterly “AI velocity” KPIs.
- Threat: Commoditization of Features & Shrinking Technical Moat (Threat Level: Medium-High)
Basic dashboards, reporting, workflows, OCR (license disks), stock management, and simple integrations become easy for anyone with APIs + LLMs. Customers expect more for the same price.
Responses / Opportunities:
- Embed Vertical AI deeply into VMG: AI-generated high-converting adverts from stock data, predictive lead scoring/maintenance alerts, natural language queries (“Show vehicles with highest margin risk”), smart anomaly detection, auto-reconciliation.
- Differentiate with SA-specific intelligence: Hyper-local compliance engine (SARS, VAT, POPIA, 2nd Hand Goods), predictive insights trained on your 500+ customers’ anonymized data, and “second brain” capabilities for dealerships.
Action: Prioritize “AI Assistant” module (e.g., VMG Voice/Genius). Use your data advantage—new entrants start from zero.
- Threat: Rising Customer Expectations & DIY Temptation (Threat Level: Medium)
Dealers may question value or slow sales cycles with “Can’t we build this with AI?” perceptions. Expectations for intelligent features rise rapidly.
Responses / Opportunities:
- Position as Intelligent Business Partner, not just software. Offer end-to-end problem-solving (insights, automation, decision support, predictive analytics for stock/pricing/recon).
- Strengthen relationships via proactive support, usage data insights, and co-creation (e.g., customer advisory board for AI features).
- Switching cost lock-in: Deep historical data value, custom configs, seamless mobile app, and unified workflows.
Action: Develop customer-facing AI demos and “AI Value Reports” showing ROI from your platform vs. DIY.
- Threat: Pricing Pressure & Perceived Lower Value (Threat Level: Medium)
If software seems cheaper/faster to build, renewals face pushback. Margins could tighten.
Responses / Opportunities:
- Tiered pricing with AI premium tiers (core DMS + AI Insights/Assistant). Move toward outcome/value-based elements where appropriate.
- Highlight total cost of ownership: Your reliability, support (“throat to choke”), compliance peace-of-mind, and integrations far outweigh cheap MVPs.
- Monetize data/insights ethically (anonymized benchmarks for the industry).
Action: Review pricing model; bundle AI features that deliver clear ROI (e.g., faster stock turnover, higher close rates).
- Threat: Internal Development & Talent Challenges (Threat Level: Ongoing)
Mid-level devs in “danger zone”; need for new skills in architecture, specification, AI oversight, validation. Risk of inconsistent AI use (hallucinations, security, IP).
Prescribed Responses (Your Mentioned Ideas + More):
- AI Use Policy (critical—draft immediately): Approved tools (e.g., Claude Enterprise, GitHub Copilot Business, vetted local models), data privacy rules (never input customer/PII/sensitive code), mandatory human review of outputs, security/compliance alignment (POPIA), acceptable use cases, and escalation for new tools.
- Skills.md files per team/tech stack (e.g., .NET/JavaScript + LangChain/LangGraph, prompt engineering, agent orchestration, TDD for AI). Include “AI Supervision” competencies.
- Focused memory/context & guard rails**: Use tools like Sand Castle (per Pocock), PRDs/destination documents, state machines/decision tables for specs, sub-agents for isolation, rigorous code review/QA loops, and “grilling” sessions. Schedule architecture review time to retain tribal knowledge.
- Hiring/focus: Prioritize seniors strong in architecture + validation; juniors thrive with AI. Train on supervisory role.
Action for meeting with Mike B/dev team: Workshop the policy, update Skills.md, prototype an internal agentic workflow for a feature, and define guardrails (e.g., no prod deploys without review).
- Additional Recommendations for Leadership & Growth
- Double down on local moats: Compliance expertise, integrations, support, and trust. Become the “go-to” for SA regulatory updates and training.
- Become the Platform: API-first mindset + ecosystem.
- Marketing/Branding: Position as “The AI-Powered, SA-Native Intelligent DMS” — most admired through reliability + innovation. Share case studies, ROI metrics, and thought leadership.
- Valuation Impact: Proactive AI integration can *increase* valuation by showing defensibility and growth (AI features as moat/upsell). Ignoring it risks erosion.
- Quality Videos & Content for Your Team
- Your two videos (Axel Molist & Matt Pocock) — excellent starting point for workflows.
- AGENTIC WORKFLOWS: Build & Sell AI Automations (2026)” — practical business application.
- AI Agents Full Course 2026: Master Agentic AI” — broad fundamentals + multi-agent.
- Industry-specific: Search for “AI dealership inventory management” or Lightspeed/Pinewood content on AI in DMS.
- Next Steps for Meeting with Dev Team:
- Review this document + the 3 LLM responses.
- Draft AI Use Policy + Skills.md templates.
- Map 2–3 high-impact AI features (e.g., ad generator, predictive alerts) using the Pocock framework.
- Set 30/60/90-day goals for internal AI adoption and one customer-facing AI pilot.
- Schedule follow-up on platform/API strategy.
The successful implementation of this project will help position VMG not just to survive, but to lead and be admired as the innovative, trustworthy, locally expert partner. With our customer base and domain depth, embracing AI aggressively turns the threat into a decisive advantage.
TDD for AI
TDD for AI (or TDD with AI / Agentic AI) refers to applying Test-Driven Development principles in workflows where Large Language Models (LLMs), AI coding agents (like Claude Code, Cursor, etc.), or autonomous agents generate or assist with code.
Classic TDD Recap (Red-Green-Refactor)
- Red — Write a failing test that defines the desired behaviour.
- Green — Write the minimum code to make the test pass.
- Refactor — Improve the code while ensuring tests still pass.
This creates a tight feedback loop, encourages simple design, and builds a safety net of tests.
How TDD Adapts to AI/LLM Coding
In traditional coding, you write the code. With AI, the AI often writes the implementation (sometimes even helps write tests). TDD becomes a powerful guardrail and feedback mechanism for AI agents because:
- AI agents can hallucinate, introduce regressions, or over-engineer.
- Tests provide clear, executable specifications that keep the AI on track.
- It creates fast, automated feedback loops ideal for autonomous agents.
In Matt Pocock’s workshop (the video you shared), TDD is highlighted as essential for reliable autonomous AI coding loops:
- Agents work on small vertical slices.
- They write/run tests → implement code → verify → commit.
- This prevents “yes-man” loops or drifting off-spec in production features.
Common Ways Teams Apply TDD for AI
| Approach | Description | Who Does What |
| Human-led TDD | You (or a senior) write the test first, then prompt AI to implement code that passes it. | Human: Tests + oversight AI: Implementation |
| AI-assisted TDD | AI helps generate the initial tests from a spec, then implements. Human reviews/refactors. | AI: Tests + Code Human: Validation |
| Agentic/Autonomous TDD | Full agent loop: Picks task → writes tests → implements → runs tests → fixes failures → commits. | AI Agent: Whole cycle Human: Architecture + final review |
| Test-First Prompting | Prompt the AI with tests upfront so it generates code that satisfies them. | Improves security, correctness, and reduces hallucinations. |
Why It’s Especially Valuable for VMG Software
- Reduces AI risks — Hallucinations, edge cases in SA compliance (VAT 264, etc.), integrations.
- Maintains quality at 5–10× speed.
- Keeps tribal knowledge — Tests document behaviour.
- Supports sub-agents and autonomous workflows from the Pocock video.
- Excellent for your Skills.md and AI Use Policy: Mandate TDD for AI-generated code.
Practical Tip for our Team: Start simple — write a failing unit/integration test for a new feature (or have AI propose one), feed it to your agent with “Make this test pass with clean, maintainable code,” then review and refactor. Tools like pytest (Python), xUnit/.NET testing frameworks, or built-in agent skills (Matt Pocock’s TDD skill) make this smooth.
This is one of the most recommended practices in 2026 for professional AI-assisted development. It turns AI from a “fast but risky” tool into a reliable accelerator. Let me know if you want examples, prompts, or a template for your dev meeting!