What the composable-stack shape closes and the three-layer pattern the engineering team grades against
The AI-coding-tool consolidation thesis that anchored 2025 — one tool wins, the others fade, the team standardizes on a single agentic-coding surface inside the next two cycles — is no longer the shape the production-engineering team's coding-tool spend lands inside. The shape that took over inside the first week of April 2026 is the composable coding-stack pattern: Cursor 3 ships the rebuilt orchestration interface for parallel agents, OpenAI publishes an official plugin that runs inside Anthropic's Claude Code, and early adopters start running all three together — with each vendor staking a distinct layer of the stack rather than fighting for the same per-tool slot.
The operationally important shape:
- The composable-stack pattern is a three-layer substrate, not a per-tool consolidation race. The three layers the pattern crystallized into across Q2 2026 are the orchestration layer (Cursor 3 — the parallel-agent dispatcher, the per-worktree isolation, the per-task verification contract, the
/best-of-ncross-model bake-off command), the execution layer (Claude Code and Codex — the per-agent execution surface, the per-tool plugin host, the per-subagent fan-out), and the verification layer (cross-provider review — the per-output reviewer, the per-diff sign-off, the per-PR adversarial-verify pass). The pattern is composable because each layer accepts plugins from rival providers at the layer boundary; the orchestration layer's per-agent dispatcher is model-agnostic, the execution layer's plugin host is provider-agnostic, and the verification layer's reviewer is cross-provider by design. - The 2.3-tool average in the production-engineering survey is the substrate-adoption signal, not an anomaly. The Q2 2026 survey of production-engineering teams measured an average of 2.3 AI coding tools per experienced developer. The 2.3 number is not the noise around the consolidation-thesis mean; it is the substrate-adoption signal for the composable-stack pattern. The honest reading is the team that grades the per-developer productivity surface against the 2.3-tool composable stack out-translates the team that grades against the per-tool consolidation thesis inside one quarter — the per-developer-per-day throughput surface runs against the composable-stack pattern, not against the single-tool standardization.
- The per-layer competition pattern is a stable-equilibrium read, not a transient consolidation gap. Cursor competes Cursor against Antigravity 2.0 and Windsurf 2.0 at the orchestration layer, Claude Code competes Claude Code against Codex and Devin Local at the execution layer, and the verification layer is opening to cross-provider entrants the standing-contract diligence surface has not yet anchored a vendor against. The per-layer competition shape is the structurally interesting pattern; the per-tool-consolidation framing the standing procurement playbook was written against is the framing the per-layer competition pattern has already replaced.
- The orchestration layer is the load-bearing FY27 procurement decision, not the execution layer. Cursor 3's
/best-of-ncommand dispatches the same per-task workload to multiple execution-layer models in isolated worktrees and grades the outcomes against the per-task verification contract; the per-task model-selection decision is now an orchestration-layer infrastructure decision driven by workload characteristics, not an execution-layer brand-loyalty decision. The team that procures the orchestration layer against the per-team workflow-export filter, the per-worktree isolation surface, and the per-task verification contract slot — and treats the execution layer as the per-task routable surface the orchestration layer dispatches against — buys the per-team optionality the per-cycle execution-layer substrate-shift makes operationally valuable.
The structural read isn't Cursor and Claude Code and Codex are merging into one tool. It is that the composable-stack pattern stabilizes the per-tool coding-agent surface into a three-layer substrate — orchestration / execution / verification — and the FY27 procurement decision the engineering team grades against is the per-layer standing-contract decision, not the per-tool consolidation decision the playbook was written against six months ago.
What the three-layer composable stack restructures about FY27 coding-tool procurement
Four concrete shifts that follow when the per-tool consolidation thesis is replaced by the per-layer composable-stack pattern.
The coding-tool procurement line item splits into three per-layer slots, with the orchestration-layer slot anchoring the standing contract. Twelve months ago, the engineering team's coding-tool procurement line item was a single slot — which IDE-with-AI does the team standardize on. The composable-stack pattern forces the procurement spreadsheet to split the slot into three: the orchestration-layer slot (Cursor 3 anchored, Antigravity 2.0 second-source), the execution-layer slot (Claude Code and Codex dual-anchored, Devin Local as the background-agent substrate), and the verification-layer slot (cross-provider review — the slot the standing diligence surface has not yet anchored against). The orchestration-layer slot is the load-bearing standing-contract anchor because the orchestration layer dispatches against the execution layer and grades against the verification layer; the per-layer split is the artifact the FY27 plan has to encode against, and the orchestration-layer anchor is the per-layer decision the standing contract carries the longest cycle-cadence against.
The per-team workflow-export filter on the orchestration-layer slot becomes the load-bearing portability artifact. The composable-stack pattern is only portable if the per-team workflows it runs against — the per-task verification contracts, the per-worktree isolation policies, the per-PR review-cadence cycles — can be exported across the orchestration-layer surfaces. The standing procurement playbook needs the per-team workflow-export filter as a first-class diligence input: which orchestration-layer surfaces export the team's per-task verification contract, which export the per-worktree isolation policy, which export the per-PR review cadence. The team that procures the orchestration-layer slot without the workflow-export filter on the diligence input buys the per-vendor lock-in against the per-tool composable substrate the team was supposed to keep portable; the team that procures with the filter explicit buys the per-layer portability the per-cycle substrate-shift makes operationally cheap.
The execution-layer slot becomes dual-anchored against the per-cycle frontier-model cadence, not single-anchored against per-tool brand. The execution-layer substrate is now the per-task routable surface the orchestration layer dispatches against, with Claude Code and Codex carrying the load-bearing duopoly inside the FY27 cycle. The standing-contract shape against the execution-layer slot is dual-vendor anchored (Claude Code primary, Codex secondary, or the inverse) with the per-task routing decision running against the per-prompt routing policy the team's coding-agent loop maintains. The team that single-anchors the execution-layer slot against either vendor pays the per-vendor frontier-model cadence-slip risk; the team that dual-anchors buys the per-task optionality the orchestration-layer /best-of-n substrate runs the per-task model-selection decision against.
The verification layer is the standing-vendor slot the regulated-industry buyer has not yet anchored against, and the procurement function gets first-mover diligence on the slot inside the next two quarters. The verification layer — cross-provider review, per-output adversarial-verify, per-diff sign-off — is the load-bearing trust-signal the regulated-industry buyer's compliance committee underwrites the production deployment against, and the standing-vendor map at the verification layer is the slot without an incumbent inside the FY27 cycle. The team that runs the per-layer diligence on the verification layer this quarter buys the first-mover standing-contract anchor against the load-bearing trust slot the per-output verifier-coverage-gap surface grades against; the team that defers the diligence ships the per-quarter trust-signal gap the regulated-industry buyer's compliance review surfaces against the FY27 production-deployment plan.
Where the composable-stack pattern is signal and where it is noise
Four honest reads on what the composable-stack shape actually tells the buyer at the FY27 coding-tool diligence review.
Signal: the per-layer-stable-equilibrium pattern is the structurally interesting read, not the per-tool feature comparison. The standing per-tool comparison chart that ranks Cursor against Claude Code against Codex against Copilot on per-feature parity is the chart the per-layer-stable-equilibrium pattern has already replaced. The signal the FY27 plan should encode is the per-layer competition pattern — Cursor at the orchestration layer, Claude Code and Codex at the execution layer, the verification layer open to first-mover entrants — and the procurement decision the substrate requires is the per-layer standing-contract decision, not the per-tool comparison-grid decision the standing playbook was written against.
Signal: the 2.3-tool per-developer average is the substrate-adoption signal the per-developer-productivity surface grades against. The 2.3-tool number is the production-engineering team's revealed-preference signal against the per-tool consolidation thesis. The team that grades against the per-developer-productivity surface and procures the per-layer composable-stack pattern translates the 2.3-tool revealed preference into the per-week-throughput surface the per-tool standardization could not produce; the team that procures against the per-tool consolidation thesis pays the per-developer-productivity-surface gap the 2.3-tool revealed preference has already surfaced.
Noise: the per-vendor M&A-and-acquisition-cadence cycle is not a per-layer disqualifier. The per-vendor M&A-and-acquisition-cadence inside the prior two cycles — SpaceX acquiring Anysphere (Cursor), OpenAI acquiring Ona (Gitpod) — is the per-vendor signal the standing-contract negotiation has to encode against. The honest read is the per-vendor M&A is a per-cycle cadence-cap input the standing contract grades against, not a per-layer disqualifier that retires the composable-stack pattern. The per-layer substrate pattern is portable across the per-vendor M&A cadence if the per-team workflow-export filter is written portably; the team that reads the per-vendor M&A as a substrate disqualifier misreads the per-layer portability question and reverts to the per-tool consolidation thesis the substrate has already replaced.
Noise: the per-tool feature-parity announcements inside the per-cycle release cadence are not the per-layer procurement-decision input. Every per-cycle release announces per-tool feature-parity claims against the rival tools at the same layer; the per-feature-parity-announcement is the per-tool marketing surface, not the per-layer procurement-decision input. The procurement decision the FY27 plan grades against is the per-layer standing-contract decision against the per-team workflow-export filter, the per-worktree isolation surface, the per-task verification contract slot, and the per-output verifier-coverage-gap artifact — not the per-feature-parity-announcement against the rival tool at the same layer.
What the engineering team should do inside the next quarter
Four concrete actions that close the gap between the composable-stack pattern and the per-layer FY27 standing-contract decision the substrate requires.
Run a per-layer diligence sprint against the three-layer composable-stack pattern inside the next sprint. For each of the three layers — orchestration, execution, verification — run the per-layer diligence input: the per-layer standing-vendor map, the per-vendor cadence-cycle-cap, the per-team workflow-export filter, the per-task verification contract surface. The artifact the sprint produces is the per-layer diligence-input map the FY27 standing-contract negotiation grades against; the team that runs the sprint this quarter writes the per-layer standing-contract decision against the per-layer diligence input, and the team that defers the sprint writes the per-layer decision against the per-tool consolidation-thesis playbook the substrate has already replaced.
Anchor the orchestration-layer standing-contract slot against the per-team workflow-export filter, not against the per-tool feature-parity announcement. The orchestration-layer slot is the load-bearing standing-contract anchor. Anchor the slot against the per-team workflow-export filter (per-task verification contract surface, per-worktree isolation policy, per-PR review cadence cycle) the standing contract requires the orchestration-layer surface to export across; the per-team workflow-export filter is the diligence input the orchestration-layer-slot decision grades against, not the per-tool feature-parity announcement against the rival orchestration-layer surface.
Dual-anchor the execution-layer slot against Claude Code and Codex with the per-task routing policy written portably. The execution-layer slot is the per-task routable surface the orchestration-layer dispatcher runs the per-task model-selection decision against. Dual-anchor the slot against Claude Code and Codex inside the FY27 cycle, write the per-task routing policy against the per-prompt routing policy the team's coding-agent loop maintains, and ship the per-task verification contract artifact the per-task routing policy grades against. The team that dual-anchors the execution-layer slot translates the per-task routing optionality into the per-week-throughput surface the per-tool single-anchor could not produce; the team that single-anchors pays the per-vendor frontier-model cadence-slip risk the per-cycle substrate-shift surface has already surfaced against the prior two cycles.
Run the per-layer-verification slot diligence inside the next quarter against the regulated-industry-buyer trust-signal surface. The verification layer is the standing-vendor slot the regulated-industry buyer has not yet anchored against. Run the per-layer diligence on the verification slot — per-output adversarial-verify substrate, per-diff sign-off cadence, per-PR review pass — against the regulated-industry buyer's compliance review surface inside the next quarter. The team that runs the diligence this quarter buys the first-mover standing-contract anchor against the load-bearing trust slot; the team that defers ships the per-quarter trust-signal gap the regulated-industry buyer's compliance review surfaces against the FY27 production-deployment plan.
The senior-judgment work the composable-stack pattern makes operationally cheap but does not replace
The composable-stack pattern compresses the cost of picking a single per-tool consolidation winner, locking the per-team workflow against a single per-tool surface, paying the per-vendor M&A-and-acquisition-cadence cost on a single-anchor standing contract, and grading the per-developer productivity surface against the per-tool feature-parity announcement. It does not compress the senior judgment of deciding which orchestration-layer surface the per-team workflow-export filter grades against, writing the per-task verification contract the per-layer composable substrate runs against, owning the per-vendor portability envelope the FY27 standing contract underwrites, and running the per-layer-verification slot diligence against the regulated-industry buyer's compliance review surface. The teams that confuse the cheapened per-tool consolidation decision for the cheapened judgment are the teams that procure the per-layer slot against the per-feature-parity announcement and read the per-quarter standing-contract-revision post-mortem on the per-vendor lock-in the per-team workflow-export filter would have surfaced. The teams that keep the senior judgment at the center of the per-layer standing-contract decision are the teams that translate the per-layer composable-stack pattern into the per-week-throughput surface the per-tool consolidation thesis could not produce. The substrate is the leverage; the senior judgment is the load-bearing wall.
The coding-tool procurement question is no longer which AI coding tool does the team standardize on; it is which per-layer standing-contract decision the FY27 plan encodes against the three-layer composable substrate, which per-team workflow-export filter the orchestration-layer slot grades against, which per-task routing policy the execution-layer dual-anchor runs against, and which first-mover diligence the verification-layer slot lands inside the next quarter. The teams that ask the right question this quarter translate the per-layer composable substrate into the per-week throughput surface; the teams that ask the wrong one ship the per-cycle gap to the FY28 procurement review against a per-tool consolidation thesis the substrate has already replaced.
At SONNET CODE we run the AI Development engagement against the per-layer composable-stack substrate — orchestration-layer per-team workflow-export filters, execution-layer per-task routing policies, and verification-layer per-output adversarial-verify cadences. If your team is still procuring against a per-tool consolidation thesis, schedule a call — we'll walk you through the per-layer standing-contract diligence we ship inside one sprint.
