Case Study

BrainMo

BrainMo builds systems to reduce the administrative and cognitive load on primary school teachers in England. Their small engineering team was shipping fast with AI-assisted tools but struggling with a high bug rate, coordination overhead, and inconsistent output quality.

Industry

EdTech / Education

Service Type

AI Workflow Audit & Technical Consulting

Duration

One week (Feb 23–27, 2026)

Result

40–50% development time saved

The Challenge

BrainMo's engineering team was using AI-assisted development tools but consistently running into the same categories of bugs:

  • 43% UI bugs — visual details missed or not implemented to spec
  • 30% feedback-loop bugs — developers not following specifications exactly
  • High regression rate — fixes breaking existing functionality
  • 40% coordination overhead — too much synchronous pairing required to catch issues

The team had already done a thorough internal review of their own processes. They understood the pain points. What they needed was an external technical perspective: what tools, processes, and automation would specifically address those patterns.

What We Did

Tobias conducted a one-week technical workflow and tech stack review:

  1. Kick-off call — 60-minute overview with the product and engineering leadership
  2. Engineer interviews — two 30-minute sessions with the senior developers to understand their day-to-day workflow, tooling, and friction points
  3. Async review — analysis of their Miro architecture documentation, Figma design files, and development workflow
  4. Written recommendations report — a prioritised, actionable document covering tooling decisions, workflow improvements, root cause analysis, and an implementation roadmap
  5. 60-minute walkthrough — live session with founders and engineering managers to walk through every recommendation

Key recommendations delivered:

  • Switch primary AI coding tool to Codex App (GPT Codex) for better spec adherence and UI accuracy
  • Implement a three-stage code review loop: local Codex review → Codex CLI with high-reasoning model → human review
  • Activate existing Playwright setup for visual regression testing
  • Set up Sentry for production error monitoring (had none)
  • Upgrade Vercel to Pro to enable deployment protections and CI/CD gates
  • Consolidate specs into a single source of truth alongside the codebase
  • Define a spec-first development process with designer-created feature documents before implementation begins

The Result

Three weeks after implementation, Kerry Hugill reported back:

"We've now been able to put the new workflow and tools through more of a wringer, and it looks like it's saving us 40–50% development time, with greatly improved accuracy!"

The team adopted both the tooling changes and the spec-first workflow. The Design Spec Docs — previously missing — added some upfront time but made the downstream implementation significantly more accurate and predictable.

The development and design teams aligned on a new handover process, reducing the back-and-forth that had accounted for nearly 40% of coordination time.

Build with us. Reach out today.

Shipping fast with AI tools but fighting bugs? A vibe coding audit finds the root causes and gives you a systematic fix.