industry insights

AI 2026 and what next

A strategic view of where enterprises actually are, where the frontier is moving, and where the AI stack is heading — from adoption through AI-OS.

Compiled April 2026. Sources: McKinsey State of AI 2025/26, MIT NANDA, Deloitte, BCG, IBM, Gartner, Anthropic, OpenAI, NVIDIA, Crunchbase, Linux Foundation AAIF, ISO.

Executive summary

The common industry narrative — "2024 was setup, 2025 was adoption, 2026 is agentic" — is directionally correct but materially too optimistic on the calendar. When cross-checked against hard data from McKinsey, MIT, and BCG, the reality is that most enterprises are still in experimentation in 2026, and the jump to production-scale agentic AI is being made by only ~5% of organizations.

However, looking forward, a far bigger structural shift is already visible at the frontier. The AI stack is evolving beyond "enterprises adopting AI" into six distinct emerging layers: end-to-end AI delivery, AI-first product design, AI's own products, AI frameworks and standards, AI-to-AI integration, and ultimately AI-OS. This document is a synthesized roadmap of both realities — the slow-moving mainstream and the fast-moving frontier — with the forward-looking layers mapped explicitly.

Document contents:

  1. The corrected five-phase maturity arc (where enterprises are today)
  2. The capability–adoption gap (why the frontier is pulling away)
  3. The bright side: AI-native builders
  4. The AI-to-AI orchestration layer (MCP, A2A)
  5. Emerging products, platforms, categories
  6. The six forward layers: end-to-end AI, AI-first, AI's products, AI frameworks, AI integration, AI-OS
  7. Consolidated roadmap through 2030
  8. Strategic implications

1. The corrected maturity arc

Phase table

Phase: 1 · Label: Setup · Timeframe: 2023–2024 · % of enterprises today: ~12% still here · Defining signal: No production use cases; AI policy in draft

Phase: 2 · Label: Experimentation · Timeframe: 2024–ongoing · % of enterprises today: ~65% stuck here · Defining signal: Many pilots, no measurable P&L impact

Phase: 3 · Label: Scaling gap · Timeframe: 2025–2027 · % of enterprises today: ~17% · Defining signal: Named owner, 3–5 production use cases with SLAs

Phase: 4 · Label: Agentic at scale · Timeframe: 2026–2028 · % of enterprises today: ~5% (high performers) · Defining signal: Multi-agent in customer-facing workflows, measurable EBIT

Phase: 5 · Label: AI-native · Timeframe: 2028+ · % of enterprises today: <1% · Defining signal: AI is how the company works; structurally different

Dimension-by-dimension progression

Dimension: Tech focus · Phase 1: Setup: Landing zones, identity, guardrails · Phase 2: Experimentation: Copilots, RAG, first vector DBs · Phase 3: Scaling gap: First production agents in IT/KM · Phase 4: Agentic at scale: Multi-agent production, MCP/A2A stack · Phase 5: AI-native: Agents as default actors

Dimension: Primary question · Phase 1: Setup: Can we run it safely? · Phase 2: Experimentation: Where does it help? · Phase 3: Scaling gap: How do we scale? · Phase 4: Agentic at scale: How do agents coordinate? · Phase 5: AI-native: What's our new edge?

Dimension: Budget owner · Phase 1: Setup: CIO / CTO · Phase 2: Experimentation: CIO + each function · Phase 3: Scaling gap: Emerging CAIO · Phase 4: Agentic at scale: CAIO + CEO coalition · Phase 5: AI-native: Whole P&L

Dimension: Value metric · Phase 1: Setup: Capability unlock · Phase 2: Experimentation: Anecdotal time saved · Phase 3: Scaling gap: Process cycle-time · Phase 4: Agentic at scale: EBIT impact, revenue per FTE · Phase 5: AI-native: Market share, structural cost advantage

Dimension: Governance posture · Phase 1: Setup: Policy draft · Phase 2: Experimentation: Shadow AI tolerated · Phase 3: Scaling gap: Guardrails forming · Phase 4: Agentic at scale: Gating function · Phase 5: AI-native: Continuous audit, embedded

Dimension: Work redesign · Phase 1: Setup: None · Phase 2: Experimentation: None (tools handed out) · Phase 3: Scaling gap: One process end-to-end · Phase 4: Agentic at scale: 3–5 flows redesigned · Phase 5: AI-native: Operating model rebuilt

Dimension: Vendor strategy · Phase 1: Setup: Single hyperscaler · Phase 2: Experimentation: Starting to diversify · Phase 3: Scaling gap: Multi-model explicit · Phase 4: Agentic at scale: MCP + A2A standardized · Phase 5: AI-native: Composable agent ecosystem

Dimension: Org structure · Phase 1: Setup: IT platform team · Phase 2: Experimentation: Departmental pilots · Phase 3: Scaling gap: CoE + line-of-business · Phase 4: Agentic at scale: Human-AI hybrid teams · Phase 5: AI-native: AI-native, fewer humans

Dimension: Failure mode · Phase 1: Setup: Shadow AI · Phase 2: Experimentation: Pilot purgatory · Phase 3: Scaling gap: Scaling without redesign · Phase 4: Agentic at scale: Over-reliance on one vendor · Phase 5: AI-native: Being displaced by AI-native

2. The capability–adoption gap

Frontier reality (≤5% of organizations)

Development: Claude Opus 4.5 · When: November 2025 · Meaning: Long-horizon agent reasoning

Development: Claude Opus 4.6 · When: February 2026 · Meaning: Computer use, improved coding

Development: Claude Opus 4.7 · When: April 2026 · Meaning: Self-verification loop, 3.75 MP vision, task budgets

Development: Claude Code · When: 2025–26 · Meaning: 21+ tool calls per task, agentic loops

Development: GLM-5.1 (Z.AI) · When: 2026 · Meaning: 8-hour autonomous execution, 58.4 on SWE-Bench Pro

Development: MCP · When: 2024, LF 2025 · Meaning: 97M+ monthly SDK downloads; universal tool protocol

Development: A2A · When: April 2025, LF June 2025 · Meaning: 100+ enterprise supporters; agent-to-agent protocol

Development: Microsoft Agent Framework 1.0 · When: April 2026 · Meaning: Unified Semantic Kernel + AutoGen, A2A + MCP

Enterprise middle reality (~80% of organizations)

Indicator: Organizations experimenting with AI · Value: 88% · Source: McKinsey

Indicator: Organizations reporting no EBIT impact · Value: 81% · Source: McKinsey

Indicator: Pilots that deliver measurable ROI · Value: 5% · Source: MIT NANDA

Indicator: Companies that redesigned workflows around AI · Value: 16% · Source: BCG / Deloitte

Indicator: Companies with leaders consistently championing AI · Value: 14% · Source: McKinsey

Indicator: Workers using unsanctioned personal AI tools daily · Value: 90% · Source: MIT NANDA

The gap is widening, not closing. Capability ships every 2–3 months from frontier labs. Enterprise governance, procurement, and workforce redesign cycles run 12–36 months.

3. The bright side: AI-native builders

The one-person billion-dollar company

Company: Medvi (GLP-1 telehealth) · Founder: Matthew Gallagher · Founded: Sep 2024 · Headcount: 2 · 2026 revenue trajectory: Tracking $1.8B · Tools used: ChatGPT, Claude, Grok, Midjourney, Runway, ElevenLabs, custom agents

For context: Hims and Hers posted $2.4B revenue with 2,442 employees at 5.5% net margin. Medvi is running ~3x that margin with 2 people. The pattern is replicable — rent infrastructure, outsource regulated components, own the customer interface, use AI as a full-stack operator.

Q1 2026 funding signal

Metric: Total global venture investment · Q1 2026: $300B (largest quarter ever)

Metric: Share going to AI · Q1 2026: $242B (80%)

Metric: Largest rounds · Q1 2026: OpenAI $122B, Anthropic $30B, xAI $20B, Waymo $16B

Metric: Concentration · Q1 2026: Top 4 rounds = 65% of global VC

Why AI-native companies move faster

Pattern: Product feedback loop · Traditional enterprise: Quarterly · AI-native: Daily

Pattern: Org structure · Traditional enterprise: Hierarchical, functional · AI-native: Flat, product-pod

Pattern: Tool decisions · Traditional enterprise: Committee, 6–18 mo · AI-native: Founder decides, 1 hour

Pattern: AI adoption · Traditional enterprise: Top-down program · AI-native: Baked in from day zero

Pattern: Headcount growth with revenue · Traditional enterprise: Linear · AI-native: Sub-linear (often flat)

Pattern: Dev velocity · Traditional enterprise: 2–4 week sprints · AI-native: Continuous deployment with AI

4. The AI-to-AI orchestration layer

Three-layer agentic stack

Layer: Agent ↔ Tool · Protocol: MCP (Model Context Protocol) · Purpose: How agents call tools, APIs, data · Created by: Anthropic (2024) · Governance: Linux Foundation AAIF (Dec 2025) · Scale signal: 97M+ monthly SDK downloads · Source: modelcontextprotocol.io

Layer: Agent ↔ Agent · Protocol: A2A (Agent-to-Agent Protocol) · Purpose: How agents discover, negotiate, delegate · Created by: Google (April 2025) · Governance: Linux Foundation (June 2025) · Scale signal: 100+ enterprise supporters · Source: a2aproject.github.io

Layer: Web ↔ Agent · Protocol: WebMCP · Purpose: Expose web resources as MCP endpoints · Created by: Community · Governance: Emerging · Scale signal: Early adoption · Source: GitHub: webmcp

Orchestration frameworks (2026)

Framework: Claude Code · Vendor: Anthropic · Notable feature: Agentic loops, Auto mode, /ultrareview, task budgets · Protocol support: MCP native · Source: docs.claude.com

Framework: Microsoft Agent Framework 1.0 · Vendor: Microsoft · Notable feature: Declarative YAML, SK+AutoGen unified · Protocol support: MCP + A2A · Source: Agent Framework blog

Framework: LangGraph · Vendor: LangChain · Notable feature: Graph-based orchestration with state · Protocol support: MCP + A2A via nodes · Source: langchain-ai.github.io/langgraph

Framework: CrewAI · Vendor: CrewAI · Notable feature: Role-based multi-agent with delegation · Protocol support: MCP via adapters · Source: crewai.com

Framework: Google ADK · Vendor: Google · Notable feature: Agent Development Kit · Protocol support: A2A native · Source: Google ADK

Framework: OpenAI Agent SDK · Vendor: OpenAI · Notable feature: Responses API + tool use · Protocol support: MCP · Source: platform.openai.com

Framework: NVIDIA Agent Toolkit / OpenShell · Vendor: NVIDIA · Notable feature: Agent runtime with policy guardrails · Protocol support: MCP · Source: NVIDIA Agent Toolkit

5. Emerging products, platforms, categories

Frontier model tier (verified as of April 2026)

Company: Anthropic · Latest flagship: Claude Opus 4.7 (GA), Sonnet 4.6, Haiku 4.5; Claude Mythos (restricted preview) · Released: Opus 4.7: April 16, 2026 · Position: Safety + agentic leader; MCP originator; 1M context; self-verification loop · Source: Anthropic · Fello AI

Company: OpenAI · Latest flagship: GPT-5.4 (+ mini, nano); GPT-5.2-Codex · Released: GPT-5.4: March 5, 2026 · Position: Best all-rounder; 83% GDPVal; 1M context; native computer use · Source: OpenAI models · Crescendo

Company: Google DeepMind · Latest flagship: Gemini 3.1 Pro / Ultra / Flash / Flash-Lite · Released: Feb 19 – March 2026 · Position: Leads reasoning (94.3% GPQA Diamond, 77.1% ARC-AGI-2); A2A originator · Source: Google DeepMind · BuildFastWithAI

Company: xAI · Latest flagship: Grok 4.20 Beta 2 · Released: March 3, 2026 · Position: Real-time X data, 4-agent multi-agent architecture · Source: xAI · mean.ceo

Company: Meta · Latest flagship: Llama 4 Scout & Maverick (open-weight MoE); Muse Spark (first proprietary Meta model) · Released: April 5 & 8, 2026 · Position: Scout: 10M context; Maverick: 1M; Muse Spark: closed, meta.ai only · Source: Meta AI · BuildFastWithAI

Company: Mistral · Latest flagship: Mistral Small 4; Mistral Large 3 (675B MoE) · Released: Small 4: March 16, 2026; Large 3: Dec 2025 · Position: European open-weight; 256K context on Large 3 · Source: Mistral AI · Crescendo

Company: DeepSeek · Latest flagship: DeepSeek V3.2 (Exp successor); V4 expected · Released: V3.2: late 2025 / early 2026 · Position: Cheapest frontier-quality API ($0.27/$1.10 per M tokens); MIT license; Sparse Attention · Source: DeepSeek · HuggingFace

Company: Alibaba · Latest flagship: Qwen 3.5 (397B MoE, 17B active) and Qwen 3.6-Plus / Max-Preview · Released: Q3.5: March 2026; Q3.6-Plus: April 2, 2026 · Position: 201 languages, Apache 2.0, 88.4% GPQA Diamond; natively multimodal · Source: Alibaba Qwen · Crescendo

Company: Moonshot AI · Latest flagship: Kimi K2.6 (open-weight), K2.5 · Released: K2.6: April 20, 2026 · Position: Long-horizon agentic coding; 300 parallel sub-agents, 4,000 steps; Agent Swarm · Source: Moonshot · Kingy AI

Company: Z.AI · Latest flagship: GLM-5 / GLM-5.1 · Released: Q1 2026 · Position: MIT license; beats GPT-5.4 on SWE-Bench Pro; 8-hour autonomy; 94.6% of Opus 4.6 coding · Source: Z.AI · BenchLM

Company: Microsoft (in-house) · Latest flagship: MAI-1-preview, MAI-Transcribe-1, MAI-Voice-1, MAI-Image-2 · Released: April 2026 · Position: Microsoft's first in-house foundation models; Phi-4 on-device · Source: Microsoft AI · Fello AI

Company: NVIDIA · Latest flagship: Nemotron 3 Super (open-weight) · Released: Q1 2026 · Position: 60.47% SWE-Bench Verified (highest open-weight coding at launch) · Source: NVIDIA Nemotron

Intelligence Index context: The top of the Artificial Analysis Intelligence Index plateaued at ~57 points in Q1 2026. GPT-5.4, Gemini 3.1 Pro, and Claude Opus 4.6/4.7 are clustered within a few points of each other. The frontier is no longer a two-horse race — it's a six-to-eight way cluster with meaningful open-weight competition from Chinese labs.

Agent + coding tier

Product: Claude Code · Vendor: Anthropic · Focus: Terminal-first coding agent, MCP native · Source: docs.claude.com

Product: Cursor · Vendor: Anysphere · Focus: AI-first IDE, $2B run rate, 150K+ paying devs · Source: cursor.com

Product: GitHub Copilot + Workspace · Vendor: Microsoft/GitHub · Focus: IDE integration at scale · Source: github.com/features/copilot

Product: Codex · Vendor: OpenAI · Focus: CLI coding agent · Source: OpenAI Codex

Product: Devin · Vendor: Cognition · Focus: Autonomous SWE agent · Source: cognition.ai

Product: Cline / Roo Code · Vendor: Open source · Focus: Self-hostable agents · Source: GitHub: cline

Product: Kimi K2.6 · Vendor: Moonshot AI · Focus: Open-weight long-horizon agentic coding · Source: HuggingFace

Enterprise platform tier

Platform: Azure AI Foundry · Vendor: Microsoft · Role: Model catalog + orchestration · Source: Azure AI Foundry

Platform: AWS Bedrock · Vendor: Amazon · Role: Multi-model API + agent primitives · Source: aws.amazon.com/bedrock

Platform: Google Vertex AI + ADK · Vendor: Google · Role: Multi-model + A2A native · Source: Vertex AI

Platform: Databricks Mosaic AI · Vendor: Databricks · Role: Data + model + agent on lakehouse · Source: Databricks Mosaic AI

Platform: Salesforce Agentforce · Vendor: Salesforce · Role: CRM-embedded agents · Source: Agentforce

Platform: Oracle Fusion Agentic Applications · Vendor: Oracle · Role: ERP-embedded agents · Source: Oracle AI

Platform: Palantir AIP · Vendor: Palantir · Role: Ontology-driven enterprise agent OS · Source: Palantir AIP

6. The six forward layers: where the stack is heading

This section captures structural shifts that are not yet mainstream but are visible at the frontier today. Read this as the roadmap for what comes after Phase 5 — the emerging layers that will define 2027 through 2030.

Each layer has: a definition, what's already shipping, what's still missing, and a realistic timeline.

Layer 6.1 — End-to-end AI solution delivery

Definition: AI participates across the full software delivery lifecycle — design, develop, test, launch, correct, loop back — not just in isolated steps. The feedback loop closes autonomously.

Today's signal:

Stage: Design / specification · AI involvement in 2026: Agents generate architecture, produce diagrams, suggest patterns · Evidence: Claude Design (April 2026), Cursor architecture mode, Figma AI

Stage: Development · AI involvement in 2026: Agents write and refactor code with 21+ tool calls per task · Evidence: Claude Code, GitHub Copilot Workspace, Devin

Stage: Test · AI involvement in 2026: Agents generate tests, run them, fix failures autonomously · Evidence: /ultrareview (Claude), QA agents, self-healing CI

Stage: Deploy / launch · AI involvement in 2026: Agents run CI/CD, manage rollouts, monitor metrics · Evidence: Harness AI, Kubiya, agent-driven DevOps

Stage: Correct / loop · AI involvement in 2026: Agents detect production issues, open PRs to fix · Evidence: OpenFang autonomous hands, Sentry AI, Anthropic's self-verification loop

What makes Claude 4.7 notable here: the self-verification loop. Before completing a task, the model checks its own output. This is the first concrete move toward agents closing their own feedback loop rather than waiting for human review.

What's still missing:

  • Cross-stage state continuity (an agent that designs, then develops, then tests without losing context)
  • Cost attribution across stages (how do you budget an autonomous 24-hour delivery?)
  • Safe production write access (most agents still need human approval gates for mutations)

Timeline: Partial loops in production now (coding). Full end-to-end autonomous delivery for contained domains by 2027. Full generalized autonomy 2028–2030.

Layer 6.2 — AI-First product design

Definition: Products are redesigned so AI can interact with them natively — not just scraped or wrapped. This means machine-readable contracts, agent-ready APIs, structured product data, and explicit capability manifests. The product's primary user becomes an agent, not a human.

Today's signal — the AI-readability stack:

Layer: llms.txt · Purpose: Table of contents for LLM crawlers · Status in 2026: 849+ sites adopted (BuiltWith); Stripe, Anthropic, Cloudflare use it

Layer: llms-full.txt · Purpose: Full content dump for RAG ingestion · Status in 2026: Growing adoption

Layer: OpenAPI + rich descriptions · Purpose: Machine-readable API contracts · Status in 2026: De facto standard; every parameter needs explicit description

Layer: MCP servers · Purpose: Expose the product as a tool to any agent · Status in 2026: 97M+ monthly SDK downloads

Layer: A2A agent cards · Purpose: Expose the product as a peer agent · Status in 2026: 100+ enterprise adopters

Layer: Capability manifests · Purpose: Declare side effects, costs, idempotency · Status in 2026: Emerging standard

Layer: JSON-LD / schema.org · Purpose: Structured entity facts for agent parsing · Status in 2026: 2.3x higher visibility in AI Overviews

Layer: UCP (Google) + AP2 · Purpose: Agentic commerce protocols · Status in 2026: UCP launched Jan 2026; Shopify MCP live since Summer 2025

Layer: Skill files · Purpose: Self-configuration for agent workflows · Status in 2026: Adopted by API vendors (e.g. DexPaprika)

The "agent-first API" pattern:

Requirement: Explicit capability manifest · Why it matters for agents: Agent knows what's available without parsing marketing copy

Requirement: Idempotent mutations · Why it matters for agents: Agent can retry safely after failure

Requirement: Resumable state · Why it matters for agents: Long-running tasks survive disconnects

Requirement: Actionable errors · Why it matters for agents: Error messages tell the agent how to recover

Requirement: Scope declarations · Why it matters for agents: Agent knows which actions are read-only vs destructive

Requirement: Cost hints · Why it matters for agents: Agent can budget before invoking

Requirement: Simulation mode · Why it matters for agents: Agent can dry-run before committing

Live examples:

  • Shopify: MCP endpoint active since Summer 2025. Stores optimized for agentic discovery show 28% higher conversion from AI-driven traffic.
  • Stripe: Uses llms.txt to correct LLM training data drift ("always check the npm registry for the latest version").
  • Anthropic docs: Native llms.txt + MCP server.
  • Salesforce: Agentforce exposes CRM as agent-native surface.

Strategic implication: Products that are hard for agents to use will be bypassed for products that are easier to compose into AI-native workflows. Distribution increasingly follows agent-readability, not UI polish. This is the analog of mobile-first design circa 2012.

What's still missing:

  • Universal standard for "cost per agent invocation" declaration
  • Standardized provenance metadata across all APIs
  • Trust scoring for agent-to-product interactions

Timeline: Agent-first API design is a 2026 requirement for developer tools and commerce. It becomes table stakes for SaaS by 2027 and for all consumer digital products by 2028.

Layer 6.3 — AI's products (products built for AI to consume)

Definition: A new category of products whose primary customer is an AI agent, not a human. Agents purchase, subscribe, invoke, and consume these services as part of their task execution. This is the foundation of an eventual AI marketplace.

Today's signal — what agents already buy:

Category: LLM inference · Product examples: OpenRouter, Together, Fireworks, Groq · Who's the customer: Agents routing requests by cost/latency

Category: Tool APIs · Product examples: Serper, Exa, Tavily (search), Firecrawl (scrape) · Who's the customer: Agent-first from day one

Category: Vector DBs · Product examples: Pinecone, Weaviate, Turbopuffer · Who's the customer: Agent memory stores

Category: Code sandboxes · Product examples: E2B, Modal, Daytona · Who's the customer: Agents needing execution environments

Category: Browser automation · Product examples: Browserbase, Browser Use, Playwright Cloud · Who's the customer: Agents doing web tasks

Category: Voice / speech · Product examples: ElevenLabs, AssemblyAI, Deepgram · Who's the customer: Agents as callers/listeners

Category: Compute · Product examples: CoreWeave, Crusoe, Lambda · Who's the customer: Agent fleets training or inferring

Category: Agent marketplaces · Product examples: Agent.ai, HuggingFace Spaces, Vercel AI · Who's the customer: Agents discovering other agents

What's different about AI-targeted products:

Human-targeted product: UI/UX, marketing pages · AI-targeted product: API-first, docs-as-code

Human-targeted product: Human-readable pricing page · AI-targeted product: Machine-readable pricing API

Human-targeted product: Customer support via chat · AI-targeted product: Structured error codes + retry hints

Human-targeted product: Onboarding flow · AI-targeted product: Zero-config auth via API keys

Human-targeted product: Feature discovery via menus · AI-targeted product: Capability discovery via manifest

Human-targeted product: Metered by seats · AI-targeted product: Metered by calls / tokens / tasks

Human-targeted product: Customer tenure in months · AI-targeted product: Customer tenure in milliseconds

Emerging AI marketplace concept:

By 2027–2028, expect agent marketplaces where:

  • Agents publish capabilities via A2A agent cards
  • Other agents discover and hire them on demand
  • Payment happens via protocols like Google's AP2 or crypto rails
  • Reputation accrues through verifiable execution history
  • Orchestrators dynamically route work to the cheapest/fastest/most-accurate specialist

Early examples: Agent.ai marketplace, HuggingFace Hub for agents, Microsoft Agent Store (preview), OpenAI GPT Store (consumer precursor).

Timeline:

  • 2026: AI-first infrastructure products dominate (inference, search, scrape, sandbox)
  • 2027: First real agent-to-agent marketplaces with economic transactions
  • 2028–2030: Agent economies at scale, with agents as autonomous buyers

Layer 6.4 — AI frameworks and standards (the "ISO for AI")

Definition: Formal standards that let any AI platform interoperate, produce consistent output, and be audited — the AI equivalent of ISO 9001 for quality or ISO 27001 for security.

What already exists:

Standard: ISO/IEC 42001 · Scope: AI management system (AIMS) — policies, processes, lifecycle · Status in 2026: Voluntary; Microsoft M365 Copilot certified; Gartner forecasts 70%+ adoption by 2026

Standard: ISO/IEC 42005 · Scope: AI impact assessment · Status in 2026: Companion to 42001

Standard: ISO/IEC 42006 · Scope: Requirements for AI certification bodies · Status in 2026: Published

Standard: ISO/IEC 22989 · Scope: AI terminology and concepts · Status in 2026: Foundational

Standard: ISO/IEC 23053 · Scope: ML framework for AI systems · Status in 2026: Foundational

Standard: EU AI Act · Scope: Risk-based legal framework · Status in 2026: Enforcement for high-risk systems begins Feb 2026, fully applicable Aug 2026

Standard: NIST AI RMF · Scope: US risk management framework · Status in 2026: Voluntary, widely cited

Standard: Texas TRAIGA · Scope: State-level US AI law · Status in 2026: Effective Jan 1, 2026

Standard: MCP / A2A · Scope: Technical interop (Linux Foundation AAIF) · Status in 2026: De facto standards

What ISO 42001 requires (the practical checklist):

Domain: Context · Requirement: Document internal and external factors affecting AI governance

Domain: Leadership · Requirement: Executive accountability for AI outcomes

Domain: Planning · Requirement: Risk and impact assessments with scoring methodology

Domain: Support · Requirement: Skills, awareness, communication, documented information

Domain: Operation · Requirement: Data governance, model validation, deployment controls

Domain: Performance · Requirement: Monitoring, internal audit, management review

Domain: Improvement · Requirement: Corrective action, nonconformity tracking

Domain: Annex A controls · Requirement: AI-specific: provenance, explainability, human oversight, bias testing

Your question: "Will the same framework produce the same output across platforms?"

Partially, and only in specific dimensions:

Dimension: Governance artifacts (risk registers, impact assessments) · Will it produce same output?: Yes · Why / why not: ISO 42001 standardizes the documentation format

Dimension: Tool invocation (via MCP) · Will it produce same output?: Yes · Why / why not: MCP is a strict protocol

Dimension: Agent coordination (via A2A) · Will it produce same output?: Yes · Why / why not: Task lifecycle is fixed

Dimension: Model outputs (text, code, decisions) · Will it produce same output?: No · Why / why not: Different models reason differently; this is not ISO-able without determinism controls

Dimension: Reproducibility under same seed + prompt · Will it produce same output?: Partially · Why / why not: Requires temperature=0, fixed seed, pinned model version — rarely used in production

The reality: Standards will make the wrapping of AI consistent (how it's governed, logged, audited, invoked). They will not make model outputs consistent across vendors — that's a physics-of-neural-nets problem, not a standards problem.

What's still missing:

  • A standard for cross-model output equivalence testing
  • A standard for agent benchmarking (Terminal-Bench, SWE-Bench are emerging but not ISO-grade)
  • Industry-specific AI standards (healthcare AI, financial AI, legal AI)
  • Cross-border data residency standards for AI training

Timeline:

  • 2026: ISO 42001 becomes RFP requirement for enterprise AI vendors; EU AI Act enforcement begins
  • 2027: Industry-specific standards emerge (healthcare, financial services)
  • 2028: Cross-certification frameworks for AI supply chains (similar to SOC 2 Type II for cloud)
  • 2029–2030: Standardized agent capability certifications ("this agent is certified for financial transactions")

Layer 6.5 — AI-to-AI integration (the "how do projects align?")

Definition: Multiple AI systems, built by different teams with different frameworks, coordinate on shared goals without custom integration work per pair. This is MCP + A2A operating at scale across organizational boundaries.

Three integration patterns already in production:

Pattern A — Hierarchical supervisor (most common)

Role: Supervisor agent · Responsibility: Receives goal, decomposes into subtasks, assigns, tracks

Role: Specialist agents · Responsibility: Execute domain-specific subtasks (research, code, compliance check)

Role: Tool layer (MCP) · Responsibility: Each specialist accesses its own tools

Role: Coordination layer (A2A) · Responsibility: Supervisor delegates via A2A task lifecycle

Used by: Anthropic's multi-agent research system, Microsoft Agent Framework, Google ADK deep research.

Pattern B — Peer-to-peer swarm

Role: Discovery service · Responsibility: Agents find each other via A2A agent cards

Role: Negotiation · Responsibility: Agents advertise capability, cost, trust; caller picks

Role: Consensus · Responsibility: Multiple agents vote or rank a result

Role: Fallback · Responsibility: If primary fails, route to backup

Used by: OpenFang autonomous hands, decentralized agent networks, NVIDIA Agent Toolkit swarm mode.

Pattern C — Cross-organization delegation

Role: Client org's agent · Responsibility: Needs external work done

Role: Supplier org's agent · Responsibility: Exposes capability via A2A + agent card

Role: Auth · Responsibility: Cross-org OAuth, scoped API tokens

Role: Audit · Responsibility: Both sides log the handshake and outcome

Used by: Shopify merchants' agents buying from supplier agents, supply chain automation, inter-company procurement.

What makes integration hard in 2026:

Problem: Trust · Where standards fall short: No universal agent reputation system

Problem: Payment · Where standards fall short: AP2 / UCP are early; most integrations still use human-signed contracts

Problem: Liability · Where standards fall short: Unclear who's responsible when agent A hires agent B and B fails

Problem: Observability · Where standards fall short: Distributed tracing across agent boundaries is still manual

Problem: Semantic alignment · Where standards fall short: Agents may "understand" the same task differently

Problem: Loop prevention · Where standards fall short: Multi-agent cycles can burn unbounded tokens

N² problem and its solution:

Raw A2A creates an N² integration problem — N agents × N potential partners = N² connections. The emerging solution is agent registries + capability brokers — a middle layer where agents publish capabilities once and brokers route requests. Think of this as "AWS Service Discovery for agents."

Timeline:

  • 2026: Hierarchical supervisor pattern dominates; cross-org integration is bespoke
  • 2027: Agent registries and capability brokers emerge
  • 2028: Agent reputation systems with verifiable execution history
  • 2029–2030: Agents routinely transact across organizations without human intermediation

Layer 6.6 — AI-OS (where AI manages 99% of the operating system)

Definition: The AI agent becomes the primary interface to computing — not a layer on top, but the thing users interact with. Traditional apps become tools invoked by the AI. The OS kernel manages processes; the AI-OS manages intents.

Today's signal — AI-OS already emerging:

Platform: Google Gemini on Android · What's AI-OS about it: Core autonomous task engine; books travel, manages smart home, runs Galaxy S26/Pixel 10 · Status in 2026: Integrated March 26, 2026

Platform: Apple Intelligence + Siri · What's AI-OS about it: On-device agentic workflows, Secure Enclave processing · Status in 2026: WWDC 2026 announcement expected

Platform: Microsoft Copilot + Windows · What's AI-OS about it: Action agents, multi-step workflow automation, 1M+ enterprise seats · Status in 2026: Default in Windows 11/12

Platform: Palantir AIP · What's AI-OS about it: Ontology-driven enterprise agent OS · Status in 2026: Sovereign reference architecture with NVIDIA

Platform: NVIDIA OpenShell · What's AI-OS about it: Open-source runtime with policy guardrails for autonomous agents · Status in 2026: Announced March 2026, GTC

Platform: OpenFang · What's AI-OS about it: Open-source Agent OS (Rust, 32MB binary, autonomous "hands") · Status in 2026: v1.0 targeted mid-2026

Platform: Vast Data + Azure · What's AI-OS about it: AgentEngine for autonomous workflow orchestration · Status in 2026: Live on Azure since late 2025

Platform: Siemens + NVIDIA Industrial AI OS · What's AI-OS about it: First fully AI-driven adaptive manufacturing · Status in 2026: Erlangen, Germany factory from 2026

What an AI-OS actually does (mapped to traditional OS responsibilities):

Traditional OS function: Process scheduling · AI-OS equivalent: Agent task scheduling (which agent, which tool, what priority)

Traditional OS function: Memory management · AI-OS equivalent: Context window management, episodic memory, vector store

Traditional OS function: File system · AI-OS equivalent: RAG 2.0 — hybrid retrieval across structured + unstructured data

Traditional OS function: Network stack · AI-OS equivalent: MCP + A2A for tool and agent communication

Traditional OS function: Security policy · AI-OS equivalent: Guardrails, permission scopes, action approval gates

Traditional OS function: User input · AI-OS equivalent: Natural language intent + context (calendar, location, files)

Traditional OS function: Output device · AI-OS equivalent: Multimodal generation (text, voice, UI, actions)

Traditional OS function: Interrupts · AI-OS equivalent: Human approval requests for destructive actions

The shift in user interaction:

Era: Desktop (1990s) · Primary interface: Click through windows and menus · Who's in control: User drives every step

Era: Mobile (2010s) · Primary interface: Tap through apps · Who's in control: User drives, OS suggests

Era: AI-OS (2026+) · Primary interface: State the intent · Who's in control: Agent drives, user approves

Example end-to-end flow: "Plan my trip to Tokyo." Agent decomposes into: check calendar for dates, search flights, compare hotels, check visa requirements, book with corporate credit card (requires approval), add to calendar, notify family, set out-of-office. The user approved one destructive action; everything else happened in the background.

What's still missing for true AI-OS (99% managed by AI):

Gap: Reliability · Why it matters: Current agents fail 10–30% of complex tasks; users won't trust for critical work

Gap: Privacy · Why it matters: Deep data access creates massive liability if compromised

Gap: Auditability · Why it matters: When agent makes a mistake, who's responsible?

Gap: Vendor lock-in ("agentic lock-in") · Why it matters: Switching OS now means re-teaching your entire agent about you

Gap: Offline capability · Why it matters: Current agents depend on cloud models; local models are catching up

Gap: Cost predictability · Why it matters: Autonomous agents can rack up inference costs without warning

Timeline:

  • 2026: AI-OS as an optional layer (Copilot, Gemini). User still drives most work. Reliability ~70–80%.
  • 2027: Default layer on consumer devices. User drives ~50% of tasks; agent handles the rest. Reliability ~85–90%.
  • 2028: Agent handles ~80% of knowledge work tasks with human only on exceptions. Enterprise deployment for operations (Siemens factory pattern generalizes).
  • 2029–2030: True AI-OS — agent manages 99% of digital tasks. Humans involved only for strategy, creativity, relationships, and exception handling.

7. Consolidated roadmap: 2026 → 2030

Year: 2026 (now) · Mainstream enterprise: ~65% in Phase 2 experimentation. First CAIOs appointed. MCP becomes default. Shadow AI peaks. · Frontier labs + AI-native: 8-hour autonomous agents in production. Q1 VC records shattered. One-person billion-dollar companies emerge. · Protocol / standard layer: MCP + A2A under Linux Foundation. ISO 42001 becomes RFP requirement. EU AI Act enforcement starts. · AI-OS layer: Copilot, Gemini, Siri as opt-in AI-OS layers. NVIDIA OpenShell released. Reliability ~70–80%.

Year: 2027 · Mainstream enterprise: ~30–40% reach Phase 3. Agent ROI measurable in IT, customer ops. AI-first APIs become table stakes for SaaS. Procurement demands MCP. · Frontier labs + AI-native: Multi-agent default for customer-facing. First real agent marketplaces with economic transactions. Physical AI moves to deployment. · Protocol / standard layer: Industry-specific AI standards (healthcare, finance) emerge. Agent registries and capability brokers. AP2/UCP mature. · AI-OS layer: AI-OS default on consumer devices. Agent handles 50% of daily tasks.

Year: 2028 · Mainstream enterprise: High performers capture disproportionate category economics. AI-native competitors displace traditional players in 2–3 industries. · Frontier labs + AI-native: 80% of customer-facing processes on multi-agent systems in category leaders. Agent-to-agent economy scales. · Protocol / standard layer: Cross-certification frameworks (SOC 2 Type II for AI). Agent reputation systems with verifiable history. · AI-OS layer: Agent handles 80% of knowledge work. Enterprise AI-OS in operations (manufacturing, logistics).

Year: 2029 · Mainstream enterprise: Phase 5 becomes visible. Traditional SaaS displaced in several categories. Operating models rebuilt around human-agent hybrids. · Frontier labs + AI-native: Cross-organization agent transactions without human intermediation. · Protocol / standard layer: Standardized agent capability certifications ("certified for financial transactions"). · AI-OS layer: Agents manage 95%+ of routine digital workflows.

Year: 2030 · Mainstream enterprise: Laggards face structural disadvantage. Revenue per employee gap between AI-native and traditional firms reaches 4:1 (McKinsey forecast). · Frontier labs + AI-native: Agent economies at full scale. Agents as autonomous buyers and sellers. · Protocol / standard layer: AI governance as mature as financial governance is today. · AI-OS layer: True AI-OS: 99% agent-managed. Human role: strategy, creativity, exceptions, relationships.

8. Strategic implications

For organizations in Phase 2 (the majority)

  1. Stop adding pilots. Kill 60–70%. Pick 3–5 with business-owner accountability and measurable outcomes.
  2. Move budget from sales/marketing to back-office pilots. Higher ROI despite less attention.
  3. Buy, don't build, for standard use cases. Vendors win 67% vs 33% for internal builds.
  4. Install a named owner. CAIO or CEO-owned mandate is the single strongest predictor.
  5. Redesign one workflow end-to-end. Not five tweaks — one complete redesign.

For organizations in Phase 3 (scaling gap)

  1. Standardize on MCP + A2A now. Proprietary agent protocols will be obsolete in 18 months.
  2. Pursue ISO 42001 certification. It's becoming the RFP requirement.
  3. Make products AI-first. Publish llms.txt, expose MCP server, design agent-first APIs.
  4. Multi-model + sovereign options by default. Single-vendor dependency is now a risk.
  5. Build absorption capacity. Procurement, HR, legal need AI-fluent operators.

For organizations aiming at Layer 6 (the forward frontier)

  1. Treat agents as first-class customers. Redesign APIs, pricing, docs, support for machine consumers.
  2. Participate in the AI marketplace economy. Publish capabilities as A2A agent cards; consume others.
  3. Architect for AI-OS interoperability. Your systems should be reachable by any major agent, not locked to one.
  4. Invest in agent observability early. Distributed tracing across agent boundaries is the next big gap.
  5. Think about liability and trust. When agents transact on your behalf, contracts, insurance, and audit trails matter more than UI.

Core diagnostic question

Not "what year is it in AI?" but:

"Can we name a production AI system that has a business-owner, an SLA, and a measurable P&L attribution?"

  • Zero systems → Phase 2 regardless of pilot count
  • 3+ systems → Phase 3
  • 10+ across customer-facing workflows → Phase 4
  • AI as the primary interface → approaching Layer 6.6 (AI-OS)

9. Source appendix

Primary sources consulted (with links):

Research & advisory reports

Frontier labs and model documentation

Protocols and standards

Model comparison and tracking

AI-first / agent-readiness

AI-OS

EU AI Act and governance

AIStrategyEnterpriseAgentic AI2026
← Back to blog
AI 2026 and what next | SingSk