Research Note

Specification-Driven Development Market Analysis

Enterprise Platform vs. Standalone IDE: AWS Kiro vs. GitHub + Microsoft Ecosystem

Paula Silva — Software GBB Americas | Microsoft Latam GBB
March 2026

Analyst Outlook: Where SDD Is Heading (2026–2028)

The Platform Wins, Not the IDE

The AI development tooling market is consolidating around platforms, not standalone IDEs. GitHub's progression — from code hosting to CI/CD (Actions) to security (GHAS) to AI coding (Copilot) to agent orchestration (Agentic Workflows) to SDD (Spec Kit) — follows the same platform gravity that defined the cloud wars. AWS Kiro, despite its polished SDD UX, is positioned as an IDE competing against a platform. History suggests that platforms that integrate security, governance, developer experience, and deployment into a single coherent experience ultimately win enterprise adoption over best-of-breed point solutions.

Model Freedom Is Non-Negotiable

With Azure AI Foundry offering 11,000+ models, GitHub BYOK supporting AWS Bedrock, Google AI Studio, and OpenAI-compatible providers, and regulatory pressures for sovereign model deployment, single-model dependency is increasingly untenable for enterprises. Kiro's Claude-only architecture is a current structural limitation. If AWS integrates Bedrock model selection into Kiro, this gap narrows significantly — but that integration has not been announced.

SDD on the Hype Cycle

SDD is approaching the Peak of Inflated Expectations. The 78.5K stars on Spec Kit and 250K Kiro users in 3 months signal genuine developer interest. However, Fowler's critical assessment (instruction non-compliance, review burden, MDD parallel) and the Scott Logic review (10x faster without SDD for small tasks) suggest the Trough of Disillusionment awaits. The tools that survive will be those adapting to "right altitude" specification — enough structure for governance, not so much that it becomes waterfall with extra steps.

The Context Engineering Convergence

Anthropic's research on context engineering (calibrated abstraction, token budget management, just-in-time retrieval, sub-agent architectures) and GitHub's agentic memory system (cross-agent knowledge with real-time code verification) are converging toward a shared insight: the future of SDD is dynamic, not static. Specifications should be living context that adapts, not rigid documents that accumulate. This favors platform approaches integrating memory, retrieval, and verification across the development lifecycle.

Historical Parallel: MDD

Model-Driven Development promised to elevate abstraction above code in the early 2000s. It failed because models were too rigid and the abstraction was awkward. SDD risks reproducing MDD's inflexibility while adding LLM non-determinism. The differentiator may be that SDD's AI agents can adapt — but that same adaptability means they can also ignore specs entirely. The tools that solve this tension will define the category.

Prediction: Will SDD Become Standard by 2028?

Unlikely in its current form. More probable: SDD principles (explicit requirements, architecture documentation, task decomposition) will become embedded features within AI coding platforms rather than standalone workflows. Spec Kit's integration into the Copilot ecosystem already shows this trajectory. By 2028, "specification-driven" will be a capability toggle within IDEs, not a separate development methodology. We expect 40-50% of enterprise engineering organizations to use some form of spec-driven workflow for complex feature development, but fewer than 15% will apply SDD universally.

SDD on the Hype Cycle

Time → Expectations → Innovation Trigger Peak of Inflated Expectations Trough of Disillusionment Slope of Enlightenment Plateau of Productivity SDD (2026) → H2 2026

References

# Source Type
1 github.com/github/spec-kit — Spec Kit repository (78.5K stars, MIT license, 25+ agents) Open Source
2 github.com/github/gh-aw — GitHub Agentic Workflows Open Source
3 github.com/github/gh-aw-firewall — Agent Workflow Firewall (Squid proxy, egress control) Open Source
4 github.com/github/gh-aw-mcpg — MCP Gateway (DIFC guards, credential isolation) Open Source
5 github.com/github/gh-aw-actions — Agentic Workflow Actions Open Source
6 github.com/githubnext/agentics — 50+ Reusable Agentic Workflows Open Source
7 github.com/microsoft/apm — Agent Package Manager Open Source
8 github.com/microsoft/work-iq — Work IQ M365 integration (MCP server) Open Source
9 github.com/github/copilot-sdk — Multi-language Copilot SDK (TS, Python, Go, .NET, Java) Open Source
10 github.com/actions/ai-inference — AI Inference Action for GitHub Actions Open Source
11 microsoft.github.io/genaiscript — GenAIScript: programmable prompts Open Source
12 GitHub Blog — Security Architecture of Agentic Workflows Eng Blog
13 GitHub Blog — Agentic Memory System (7% PR merge improvement) Eng Blog
14 GitHub Blog — Execution Is the New Interface Eng Blog
15 GitHub Blog — Copilot CLI Practical Guide Eng Blog
16 GitHub Blog — Continuous AI for Accessibility (89% issue close rate) Eng Blog
17 GitHub Blog — Accessibility Compliance in Five Hours Eng Blog
18 GitHub Next — Continuous AI (8 use cases) Research
19 Microsoft Developer Blog — Spec Kit MS Blog
20 GitHub Docs — Copilot Coding Agent Best Practices; Vibe Coding Tutorial Docs
21 Modernize Legacy COBOL App — Brownfield SDD example Open Source
22 Anthropic — Building Effective Agents (workflows vs agents patterns) Research
23 Anthropic — Effective Context Engineering Research
24 Anthropic — Writing Tools for Agents Research
25 Claude Code — Plan Mode Docs
26 kiro.dev — AWS Kiro product page, features, pricing Product
27 Kiro Blog — General Availability (Nov 2025): PBT, checkpointing, CLI Blog
28 Kiro Enterprise — IAM, SSO, GovCloud, IP indemnity Product
29 Martin Fowler — SDD Tools Analysis — Fowler/Böckeler maturity model, MDD parallel Independent
30 Scott Logic — Spec Kit Review (10x faster without SDD) Independent
31 github.com/ruvnet/sparc — SPARC specification framework Open Source
32 github.com/ruvnet/ruflo — Ruflo multi-agent orchestration Open Source
33 Gartner — Hype Cycle for AI in Software Engineering, 2025 Analyst
34 IDC MarketScape — AI Coding and Software Engineering Technologies 2025 Analyst
35 McKinsey — "The State of AI 2025" (62% experimenting with agents, <10% at scale) Analyst
36 BCG — "AI at Work 2025" (30-40% efficiency gains; 23% dip for complex tasks) Analyst
37 Microsoft Learn — Defender for Cloud + GHAS Integration Docs
38 Microsoft Learn — Purview DSPM for AI (GA April 2026) Docs
39 GitHub Changelog — BYOK Enhancements (Jan 2026) Changelog