CASE STUDY

How Brilworks Doubled Coding Speed at Two Clients with Claude Code

Brilworks's anonymized case study on rolling out Claude Code at two product engineering teams: ~2x coding-speed lift, the four-step playbook, and where the agent does and doesn't compound.

Category: Engineering Enablement / B2B SaaS

Talk to Brilworks About Claude Code Build Something Similar

~2x

Productivity gain on agent-eligible tasks (both clients)

Clients across different industries (hospitality SaaS + smart-home)

Stage rollout: audit, agent design, enablement

How Brilworks Doubled Coding Speed at Two Clients with Claude Code

Overview

The Challenge & Vision

Two of our clients adopted Claude Code with Brilworks's help in the last two quarters. One was a hospitality / restaurant-tech SaaS we'll call HostCo; the other was a smart-home and connected-device platform we'll call HomeCo (real names withheld under NDA). Different industries, different stacks, but the same engineering reality: senior-engineer leverage was getting eaten by integration boilerplate, deferred refactors, and — at HomeCo — cross-runtime tickets touching firmware, backend, and mobile in a single change. Both teams had tried ad-hoc AI assistants; the results were uneven. They wanted leverage, not novelty. By the end of each engagement, both teams reported a roughly 2x productivity improvement on agent-eligible tasks.

Get the Full Case Study

Architecture decisions, design process, and detailed results — in a shareable PDF.

No spam — just the case study PDF.

5★ ClutchNDA Available120+ Products

Where senior engineering time was going

Across both engagements, the same patterns kept eating throughput. Different industries, different stacks — same shape of bottleneck:

Customer-specific integration scaffolding — payment gateways, POS connectors, loyalty platforms, marketing-automation hooks at HostCo: important per deal but boilerplate per ticket.
Long-deferred refactors blocked behind capacity constraints — at both clients, large refactors slipped quarter after quarter because nobody had a free week to take them on.
Test coverage gaps in older modules at HostCo — every change in those modules ran slower than it should have because of missing safety nets.
Cross-runtime tickets at HomeCo — telemetry fields, alert types, permissions touching firmware, cloud, and mobile in a single change, with domain knowledge concentrated in a few senior engineers.
Hardware compatibility constraints at HomeCo — legacy modules accumulated faster than in pure-cloud SaaS because rewrites were expensive.
No standard at either team for which tasks were worth handing to an AI assistant and which were not.

Solution

What We Built

✓
Audit and surface selection — mapped each repo to where Claude Code would compound the fastest (integration scaffolding, test generation, well-scoped refactors), deliberately excluding surfaces where the agent would create more review load than it saved.
✓
Agent design and tooling — codified team conventions (naming, error handling, test patterns, commit messages) into Claude Code instructions; wired in lint, type-check, and test runners so the agent self-verified before opening a PR.
✓
Enablement, not handoff — paired with senior engineers across two sprints, drafting prompts together, refining the instruction file, and tightening review heuristics. The team owned the agent at the end of the engagement.
✓
Multi-runtime tuning at HomeCo — instruction files per runtime (firmware, cloud, mobile) rather than per repo, with explicit go/no-go criteria for staging firmware and mobile work behind cloud-side wins.
✓
Cross-runtime ticket pattern at HomeCo — the most leveraged use of the agent turned out to be tickets where one engineer drove a change through firmware, cloud, and mobile with the agent handling the cross-cutting boilerplate.