BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.adass.org//adass2026//talk//FUWBLF
BEGIN:VTIMEZONE
TZID:AWST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1;UNTIL=20051231T160000Z
TZNAME:AWST
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
END:STANDARD
BEGIN:STANDARD
DTSTART:20070325T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:AWST
TZOFFSETFROM:+0900
TZOFFSETTO:+0800
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20071028T030000
RRULE:FREQ=YEARLY;BYDAY=4SU;BYMONTH=10;UNTIL=20081025T190000Z
TZNAME:AWDT
TZOFFSETFROM:+0800
TZOFFSETTO:+0900
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-adass2026-FUWBLF@pretalx.adass.org
DTSTART;TZID=AWST:20261103T151500
DTEND;TZID=AWST:20261103T152700
DESCRIPTION:The SKA Regional Centre Network (SRCNet) is entering its Scienc
 e Verification phase\, a period defined by tight milestones and a growing 
 surface of interdependent services — registration\, ingestion\, metadata
 \, and data access components among them — that must mature quickly with
 out compromising long-term maintainability. This contribution reports on t
 he use of AI coding assistants across SRCNet software development teams to
  meet these constraints. Rather than treating AI-generated code as a short
 cut\, our approach centers on structured prompting: requests are framed to
  explicitly encode target architecture and design patterns\, module bounda
 ries\, documentation standards\, and test coverage from the outset\, so th
 at output is usable beyond a first draft. In practice\, AI-assisted code p
 lays its strongest role in initial prototyping — standing up service ske
 letons\, exploring API contracts\, and validating design choices quickly 
 — after which ownership passes to the responsible team for integration\,
  hardening\, and long-term maintenance. This handoff surfaces a recurring 
 tension: AI tools can produce substantial\, working code far faster than a
  team can meaningfully review it\, and oversized reviews risk becoming a b
 ottleneck or a rubber stamp\, undermining the quality controls the Science
  Verification milestones depend on. We discuss practices adopted to mitiga
 te this — smaller\, prompt-scoped increments\, review checklists aligned
  to architectural intent\, and tests generated alongside code rather than 
 after it — while showing that\, managed this way\, overall development v
 elocity across SRCNet components can increase substantially. We conclude w
 ith early observations on where this approach has been most effective and 
 where human architectural judgment remains indispensable.
DTSTAMP:20261116T154859Z
LOCATION:Seminar Room 2
SUMMARY:Use of the AI for SRCNet development - Jesus Salgado
URL:https://pretalx.adass.org/adass2026/talk/FUWBLF/
END:VEVENT
END:VCALENDAR
