Customer Story
Wavect moves between client and internal repositories across product delivery, integrations, AI work, backend and frontend systems, infrastructure and ongoing maintenance. They measured what changed when LeanCTX shaped the context entering their coding agents.
The problem was not one spectacularly large prompt. It was the ordinary sequence of read, search, run, inspect, re-read, compare and verify.
No single website or stack represents the tracked experience. Wavect’s workload spans greenfield builds, inherited systems, bug investigations, integrations and internal tooling — each producing different context demands.
Different project shapes produce different context demands. Greenfield builds, inherited systems, security-sensitive backends.
Engineers repeatedly pay the orientation cost when they switch projects, stacks or delivery phases.
A routine content change and a security-sensitive backend change cannot share the same recovery and review rules.
Package managers, Git, tests, containers, crawlers and infrastructure commands all become model input.
Begin with the smallest useful representation, expand when the decision requires more evidence, and verify against the repository rather than the summary.
Find system shape, likely ownership and evidence needed for the next decision.
Narrow scope with cached context and compact representations.
Inspect exact source before exact edits. Recover raw when meaning is uncertain.
Keep signal from builds, tests and commands. Compress noise, preserve failures.
Review the diff, run the tests, check the real system. Compression does not approve work.
The savings were not evenly distributed. They concentrated in three patterns that repeat across agency delivery.
Agency work often starts before the engineer has a reliable mental model. LeanCTX delayed unnecessary detail — the full source returned when the investigation became an implementation task.
Long sessions revisit stable material for different reasons. Cached re-reads and delta-oriented views reduced the repeated payload. The value compounded with session length.
Builds, tests, Git and package managers produce more text than the next decision needs. Shell compression kept the result and actionable failures while removing noise.
LeanCTX acted more like an evidence editor than an engineer. It reduced duplication and highlighted signal, but it did not decide whether the underlying work was correct.
The case study uses 64.1% as the headline because it represents the tracked mix. Larger percentages explain where the total came from.
A credible case study makes the boundary easy to find. These percentages show context reduction — they do not automatically establish faster delivery, better code or universal product fit.
Not a controlled A/B study. The tracked workload was observational.
Not a productivity percentage. Cycle time and throughput were not measured.
Not audited invoice ROI. The 56.6% is an estimated token-cost reduction.
Not proof of better code. Smaller context does not prove correctness or security.
Not equal across formats. Compression varies between structured code and dense literal formats.
Not every agency workflow. This reflects Wavect’s specific mix, tooling and acceptance rules.
Large context alone is not enough. Repetition alone is not enough. The strongest fit appears when large repository or tool context returns throughout the same workflow.
Large, repeated workflows — long agent sessions, unfamiliar repositories, recurring reads and verbose tool output.
Small but frequent workflows — individual saving may be modest, but repetition makes the aggregate meaningful.
Large one-off workflows — compression can help, but setup, review and recovery may dominate the economics.
Small one-off tasks — a context layer can add more operational complexity than the task justifies.
LeanCTX became useful when we stopped thinking about one prompt and looked at the whole agency session. The same files, schemas and commands come back again and again. Making those repeats smaller was where the value compounded.
Run a two-week controlled evaluation with the same repository and task set. Measure context, time, quality and cost per accepted task.
Cookie preferences
We use analytics to understand how our site is used. You choose what's allowed — no tracking runs until you decide.
Essential
Site functionality, security, preferences
Analytics
Pageviews & usage patterns · PostHog EU (Frankfurt) · No cross-site tracking
Read our Privacy Policy for full details. You can change your preferences anytime.