How we built an automation workflow with AI agents in 6 weeks for £200/month
How we built an enterprise CRM with AI agents in 6 weeks for £200/month
An honest account from a 15-year UK software company on what AI coding can and can’t do in 2026.
- Built a full enterprise CRM with AI agents in 6 weeks, replacing tools costing £40k+/year.
- Six autonomous agents now manage 7,000+ contacts overnight scoring, enrichment, and follow-up.
- Total stack runs at £200/month versus £3,000+/month for enterprise SaaS.
- Custom internal tools that cost £150k+ can now be delivered for £20k–£50k in weeks.
- AI is a force multiplier, not a replacement senior engineering judgement is still essential.
If you’ve been told that AI coding tools are overhyped, this post is for you. We’ve been delivering bespoke business applications for over fifteen years and watched countless “revolutionary” tools come and go. What we’ve experienced over the last six weeks with Claude Code has fundamentally changed how we think about software delivery.
This is the honest story of how we built our own enterprise-grade CRM with AI agents and what it means for your business.
The old way of building internal tools
Until recently, businesses faced a binary choice for internal software:
For most companies, internal tools always lose budget priority to client-facing work leaving operations running on disconnected spreadsheets, email threads, and tools nobody really likes. So we did something different: we built our own, using AI as our pair programmer.
The six AI agents running our sales operation
The genuinely transformative part isn’t the CRM itself it’s what runs in the background. Six autonomous AI agents work overnight to keep the system intelligent. Every action that touches data is presented to a human for approval before execution.
The cost reality versus traditional approaches
The numbers tell a story that should make every CFO sit up:
Even accounting for the engineering time invested, the total cost of building was a fraction of what we would have paid for SaaS in the first year alone. The return on investment was positive within the first week of using the system.
More importantly, we own the system completely. When our process changes, the software changes with it.
What this means for AI coding today
Here’s the honest assessment from someone who has shipped production code for over fifteen years: AI coding tools have crossed a critical threshold.
Multi-file refactors that would take a senior engineer half a day are completed in ten minutes. Bug fixes are diagnosed and resolved with full context across the codebase. Feature implementations actually work first time, with proper error handling, type safety, and adherence to existing patterns.
But and this is the critical point it requires experienced engineers in the loop. Without architectural guidance, AI will happily build technically correct things that are strategically wrong for your business.
The honest drawbacks you need to know
This isn’t a perfect technology, and pretending otherwise would be dishonest.
- AI doesn’t know your business context, that’s a human responsibility.
- It will occasionally write incorrect code with complete confidence.
- Database operations need extra care, we lost data once during a migration.
- Costs can spiral without engineering discipline around when to use AI.
- This is not a no-code solution. You still need real engineering judgement.
What this means for your business in 2026
If you’re a business leader reading this, here’s what’s now genuinely possible that wasn’t twelve months ago. Custom CRMs, ERPs, internal portals, and operations dashboards can be built in weeks for £20k–£50k, rather than the £150k–£300k they previously required.
The window where your competitors are still using generic SaaS while you have bespoke AI-powered systems is opening right now. The companies that recognise this in 2026 will build a structural advantage that compounds over years.
The technology stack we used
| Frontend | Next.js 14, TypeScript, Tailwind CSS |
| Backend | Node.js, Python FastAPI |
| Data | PostgreSQL, Redis |
| AI | Anthropic Claude (Sonnet, Haiku, Opus) |
| Services | Apollo.io enrichment, Resend email |
| Hosting | Single Hetzner virtual server |
Frequently asked questions
Could a non-technical team build this themselves?
No. AI dramatically accelerates experienced engineers but doesn’t replace the need for them. Production grade software still requires understanding databases, security, system architecture, deployment, and monitoring.
How long would something similar take for our business?
Most internal tools we now scope sit in a 4–8 week build window depending on integrations and complexity. We’d typically scope it in a discovery call and give you a fixed-price proposal.
Who owns the code at the end?
You do. Unlike SaaS, you own the system completely when your process changes, the software changes with it. No vendor roadmap, no per-seat licensing penalties.
How do you handle data security and compliance?
We’re ISO 27001 certified and Cyber Essentials accredited. Every system we build has appropriate authentication, input validation, environment isolation, and audit logging. We can host on your infrastructure or ours, depending on your requirements.
A 30-minute discovery call where we’ll walk through the build approach and identify where AI automation could deliver the highest ROI in your specific business. No sales pitch, no obligation.