Growing One's AI Tools: A Paradigm of Practice-Grown Tools
Abstract
Professionals have long used software built for them by others. This article reports the growth of a professional computer-assisted interpreting (CAI) tool built by its own user, a conference interpreter without programming training. Drawing on an eight-week version-controlled record with telemetry from 133 logged sessions and over 20,000 recognized utterances, the article traces how real conference work drove the tool through 62 commits and six versions. The record shows capabilities deployed on the day their need appeared and one strategy change that cut median end-to-end delay by over a third within one afternoon. The article then proposes the paradigm of practice-grown tools, defined by the shift from using to growing, the co-evolution of tool and practice through breakdown and repair, bounded local computation, and an assessment corollary, and it formulates three propositions toward tool-inclusive assessment of practitioners, from tool-making literacy to a compute-standardized format for certifying what they and their tools achieve together.
Figures & tables
| Dimension | Purchased tools | Customized tools (EUD) | Practice-grown tools |
|---|---|---|---|
| Who builds | Vendor developers | User or local tailor, within a vendor design | The domain expert, with AI implementation |
| Starting point | Finished product | Finished product plus tailoring hooks | Minimum viable tool derived from the expert’s workflow |
| Growth driver | Vendor roadmap | Occasional tailoring by a minority of users | Recurrent breakdown and repair in real practice |
| Knowledge source | Market research and support channels | User feedback passed to designers | The expert’s own articulated judgment |
| Failure and repair | Vendor support queue | Customization lost on update | Repaired by the builder in the same cycle |
| Computation | Vendor infrastructure | Vendor infrastructure | Bounded local hardware owned by the expert |
| Phase | Date | Practice event | Tool response | Observed effect |
|---|---|---|---|---|
| I | Aug 27–31 | Founding-era close-out (Zhao, 2026a) | v0.2 rewrite; logging, segmentation, backend | Tool becomes its own measurement device |
| II | Sep 4 | Four-session conference day | Same-day v0.3.0 subtitle upgrades | Logging enriched by that day’s data |
| II | Sep 8 | Online-meeting scenario | v0.4.0 system-audio capture | New work mode supported |
| III | Sep 16 | Relay-work demands | Model replaced by smaller successor | Generation change post-dates founding telemetry |
| III | Sep 16–17 | Six-language relay forum; no LID | Three capability commits on Sep 17 | Capability born on the day its need appeared |
| III | Sep 19 | Flagship session; backend disconnects | v0.4.0 tag on repair commit | Version marks survival, not function |
| Session | Median | 2 s | 3 s | Cap | Rate | |
|---|---|---|---|---|---|---|
| E2E | share | (CJK/s) | ||||
| Sep 17 relay | 812 | 1,513 ms | 92.0% | 99.8% | 0% | — |
| Oct 21 AM | 1,140 | 2,461 ms | 34.5% | 84.2% | 41.8% | 4.28 |
| Oct 21 PM | 387 | 1,548 ms | 78.1% | 99.7% | 23.0% | 4.05 |
| Oct 22 | 1,097 | 2,017 ms | 49.0% | 97.1% | 46.6% | 3.87 |
Software Engineering
Software engineering studies how to design, build, test, deploy, and maintain reliable software systems and the processes behind them.
See all 290 papers