Blue Ridge mountain ridges at dawn

Software Engineering
Under Changing Constraints

How lower-cost generation creates second-order changes across people, organizations, delivery, security, and economics.

Andrey Tapekha · 2026

System overview

Generation abundance redistributes constraints.
Scarcity moves into understanding, verification, and deployment.

Cheap production shifts pressure into context, architecture, review, release capacity, security, and judgment.

clear intent & acceptancegeneration abundancechange-volume surgeparallel solution searchagentic change bundlessoftware sprawlevidence-ready generation
review-capacity gapchange impact & release safetyAI-assisted reviewattention routingindependent verificationaccountable acceptance
human judgment & learningfoundational skill decaycognitive offloadingmental-model driftautomation biastask stewardshipAI-supported learning
context, architecture & reuseshared ontology & knowledge graphstale-context riskarchitecture alignmentownership & provenancegoverned reuse
workflow redesignvariable autonomyexception ownershipdecision & stop rightscross-functional coordinationoutcome feedback loops
AI & software supply chainattack-surface expansioninsecure change at scaleprompt & tool injectionagent privilege misuseAI-enabled defensecontinuous validation
capacity & prioritiestoken & compute budgetsverification capacitydeployment capacitycost per accepted changevalue-based prioritizationworkforce rebalancing
01Generation
abundance
intent · options · volume
TODAY'S FOCUS02Verification
& trust
impact · evidence · acceptance
03Human
judgment
learn · challenge · intervene
04Org context
& reuse
ontology · architecture · reuse
05Operating
model
route · decide · recover
06Cyber
security
supply chain · integrity · defense
07Value
& capacity
tokens · review · deployment
01 / 07

Generation abundance

AI makes candidate changes cheap to produce. Review becomes the first visible bottleneck.

A queueing model of software delivery

Generation speed is not delivery speed. Every stage has its own capacity.

REVIEW QUEUE321 checks waitingREVIEW FASTER
SIZELOWGROWTHSTABLE
CHANGES → PULL REQUESTS7 SOURCES
CUSTOMER
PR-1423 commits
DEFECT
PR-1432 commits
SECURITY
PR-1445 commits
CONTROL
PR-1451 commit
DATA
PR-1464 commits
PLATFORM
PR-1472 commits
ARCH DEBT
PR-1486 commits
REVIEW QUEUE3 WAITING
TAIL · NEWEST ARRIVALPR-145
HEAD · NEXT TO REVIEWPR-143
PR-1423 commits01
PR-1432 commits02
PR-1445 commits03
PR-1451 commit04
PR-1464 commits05
PR-1472 commits06
PR-1486 commits07
PR-1493 commits08
PR-1501 commit09
PR-1514 commits10
PR-1522 commits11
PR-1535 commits12
3 PRs queued
EVIDENCE SCANSTARTING
PR-1423 commits grouped
01INTENTdesired outcome
02CHANGE FACTScode + dependencies
03BEHAVIORexpected result
04BLAST RADIUSsystem impact
05SECURITYsupply chain
06INDEPENDENT CHECKseparate evidence
07OWNER + ROLLBACKsafe to operate
VALUE DELIVERY0 ACCEPTED
MERGED✓
ACCEPTED CHANGE CAPACITY
← → POSITION · ↑ TURN · ↓ RELEASE
accepted changeready · monitored · reversible
ACTUAL FLOWcommits→pull request→review queue→seven evidence checks→merge

Research and operating agenda

Seven areas for AI-assisted software delivery

What changes

Possible responses

Recent evidence

Andrey Tapekha speaking during the industry panel at NC State's 2025 Applied AI in Engineering and Computer Science Symposium

Presentation controls

→ / ←Next / previous slide↓ / ↑Next / previous challenge on slide 4PgDnNext slide from any slidePgUpPrevious slide from any slide1–7Select a challenge on slide 4NSpeaker notesOSlide overviewFFullscreenPPrint / save as PDF?Show / hide helpEscClose overlay