Parallel Coding Agents

Complete Research Library — May 2026

10
Documents
9
Research Pillars
280
Sources Cited
Master Plan
MASTER
Parallel Coding Agents — Research Synthesis
Cross-pillar thematic synthesis with actionable recommendations
  • Bad decomp kills parallelism
  • Pre-flight detection is king
  • OCC default, lock hot spots only
  • Planner/Worker/Judge topology
  • Detect drift before acting
  • More agents worsen bad decomp
  • Coordination, not capability
  • Worktrees + merge queues rule
  • Ordered locking kills deadlocks
  • Redlock unsafe; use ZooKeeper
Research Pillars
RESEARCH
Existing Multi-Agent Coding Systems
Devin, SWE-Agent, AutoDev — how concurrent coding agents coordinate today.
  • Multi-agent coding tool survey
  • Task partitioning strategies
  • Isolation guarantees in practice
  • Coordination failure post-mortems
  • Agent communication patterns
  • Concurrent PR workflows observed
  • 30 sources cited
 RESEARCH
Git Worktree & Branch Isolation Primitives
Git worktrees, merge queues, and branch isolation as coordination primitives.
  • Worktree lifecycle management
  • Merge queue tools compared
  • Automated rebase on queue drain
  • Branch naming for agent identity
  • Worktree limitations & gotchas
  • CI integration with worktrees
  • Index lock contention patterns
  • 31 sources cited
 RESEARCH
Classical Concurrency Control Theory
2PL, MVCC, and OCC — classical CS theory applied to concurrent code edits.
  • Two-phase locking (2PL) theory
  • MVCC semantics & snapshots
  • Serializability & isolation levels
  • Intent lock hierarchies
  • Vector clocks & causality
  • Lock compatibility matrices
  • OCC theoretical foundations
  • 31 sources cited
 RESEARCH
Monorepo Coordination at Scale
How Google, Meta, and Stripe coordinate thousands of engineers in one repo.
  • Google Piper/CitC virtual FS
  • Meta Sapling & Eden FUSE FS
  • CODEOWNERS enforcement patterns
  • Pre-submit validation queues
  • Trunk-based dev at scale
  • Bazel/Buck remote execution
  • Change impact analysis tooling
  • 38 sources cited
 RESEARCH
Upfront Conflict Zone Detection
Static analysis and dependency graphs to predict agent conflict zones upfront.
  • File dependency graph construction
  • Call graph overlap analysis
  • Module boundary inference
  • Package import graph overlap
  • Change blast radius estimation
  • Task description semantic overlap
  • Pre-flight conflict check patterns
  • 32 sources cited
 RESEARCH
Lock Scope & Granularity Design
File, directory, module, or semantic — the tradeoffs in lock scope design.
  • File vs directory vs module locks
  • Hierarchical intention locks
  • Adaptive granularity under load
  • Advisory vs exclusive lock types
  • Scope declaration protocols
  • Lock coarsening vs splitting
  • Read/write lock compatibility
  • 42 sources cited
 RESEARCH
Agent Recovery After Upstream Merges
Rebase, restart, or patch — what agents do when upstream changes land.
  • Automated rebase feasibility
  • Rebase failure classification
  • Full restart vs context reuse
  • Incremental context delta replay
  • Divergence severity scoring
  • Cost model: rebase vs restart
  • Automated conflict resolution
  • 34 sources cited
 RESEARCH
Deadlock, Starvation & Fairness
Cycle detection, timeouts, and fairness policies for multi-agent lock contention.
  • Deadlock formation in agent graphs
  • Wound-wait & wait-die protocols
  • Wait-for graph cycle detection
  • Timeout + exponential backoff
  • Starvation & priority inversion
  • FIFO fairness & aging policies
  • Work-conserving scheduler design
  • 32 sources cited
 RESEARCH
Optimistic & Speculative Approaches
OCC, speculative parallelism, and CRDT-style approaches as lock alternatives.
  • OCC validate-at-commit for code
  • Speculative parallelism rollback
  • Conflict probability modeling
  • CRDT-style code merging
  • Software transactional memory
  • When OCC beats pessimistic locks
  • Empirical conflict rate thresholds
  • 10 sources cited