Prompted LinesAI guidance for insurance

Strategy · August 2026

The agent timeline, 2024–2030

From the first agent demos to fleets in production: what actually happened and when, then a clearly marked set of estimates through 2030. Past entries are sourced; future entries are reasoned guesses, and the page tells you which is which.

The timeline, at a glance

Three lanes, three clocks: model capability moves first, insurance adoption follows, governance and regulation lag by design. The widening gap between the lanes is the story. Teal entries happened and are sourced in the full timeline below; bronze entries are estimates. Scroll horizontally; select any entry to jump to its full text.

Happened (sourced) Estimate (projection)
2024 2025 2026now 2027 2028 2029 2030

Model capability

MarDevin: first autonomous software engineer SepOpenAI o1: reasoning models OctAnthropic computer use NovMCP open-sourced
JanOperator, then Deep Research FebClaude 3.7 + Claude Code MarMETR task-horizon result AprAI 2027 scenario MayClaude 4 SepSonnet 4.5: 30-hour runs
2027The lab question resolves
2028AI 2027's resolution window

Insurance adoption

OctGuidewire agrees to buy ProNavigator DecAIG forms SPV; signs Palantir
JanAllianz × Anthropic; Travelers × Anthropic FebTravelers agentic Claim Assistant AprCFC Lane Assist; Duck Creek agentic platform JulDuck Creek acquires Send AugGuidewire ships Qusar
2027Bounded agentic workflows default at leaders
2028AI-native operating models among leaders
2029Multiagent "virtual coworker" workflows mainstream
2030The two-speed industry resolves

Governance & regulation

MarNAIC exam-tool pilot begins (12 states) JulAI 2040: Plan A published NovNAIC tool expected national
DecEU AI Act high-risk obligations begin
AugEU AI Act regulated-product obligations
2029AI outcomes testing routine in most states

The timeline, in full

2024 agents become plausible

Mar

Cognition launches Devin

Billed as the first autonomous software engineer: the first widely seen demo of an agent taking a whole task rather than a single turn.1

Sep

OpenAI ships o1

Inference-time reasoning arrives in a mainstream model: it thinks before answering, the substrate agents need for multi-step work.2

Oct

Anthropic demonstrates computer use

Claude 3.5 Sonnet operates a desktop directly (screenshots, mouse, keyboard). Agents leave the chat window.3

Nov

Model Context Protocol (MCP) open-sourced

Anthropic publishes a standard way to wire models to tools and data; the industry adopts it through 2025.4

2025 agents become products

Jan

Operator, then Deep Research

OpenAI ships a consumer computer-using agent; Deep Research follows in February, producing cited research reports autonomously.56

Feb

Claude 3.7 Sonnet and Claude Code

Agentic coding enters daily developer work; the terminal agent becomes a real category.7

Mar

METR publishes the task-horizon result

The length of task agents complete autonomously is doubling roughly every seven months. The single chart this timeline leans on.8

Apr

The AI 2027 scenario

Kokotajlo, Alexander, and colleagues publish a concrete fast-takeoff path: superhuman coding agents by early 2027, then AI automating AI research. Influential, contested, and falsifiable by design.9

May

Claude 4 ships

Coding agents sustain hours-long sessions; "agentic" becomes the year's vendor word, and the sorting of real from relabeled begins.10

Sep

Claude Sonnet 4.5 runs 30+ hours autonomously

The overnight agent shift becomes routine for well-scoped engineering work.11

Oct

Guidewire agrees to acquire ProNavigator

Insurance core-system vendors start buying AI capability rather than building all of it.12

Dec

AIG forms its SPV and signs Palantir

AIG, Amwins, and Blackstone form a special-purpose vehicle alongside an AIG–Palantir GenAI collaboration; multi-agent underwriting stacks go enterprise-scale.13

2026 agents in production, and examiners

Jan

Allianz × Anthropic; Travelers × Anthropic

Allianz signs a global responsible-AI partnership; Travelers puts Claude in front of 10,000 engineers and analysts.1415

Feb

Travelers launches its agentic Claim Assistant

A production agentic voice service for claims, built with OpenAI. Agentic claims handling, in the wild.16

Mar

NAIC exam-tool pilot begins in 12 states

The AI Systems Evaluation Tool pilot runs through September. AI governance becomes an examination item, not a whitepaper topic.17

Apr

CFC's Lane Assist; Duck Creek's agentic platform

Lane Assist is billed by CFC as a world-first agentic underwriting pilot in specialty insurance (submission email to quote recommendation in seconds); Duck Creek ships agentic underwriting and claims applications.1819

Jul

Duck Creek acquires Send; AI 2040 publishes

Agentic underwriting-to-core becomes a product line. The AI Futures Project publishes AI 2040: Plan A, a policy counterweight to its own AI 2027 scenario.2021

Aug

Guidewire ships Qusar

A model-agnostic framework for building, running, and governing AI agents against core insurance systems.22

You are here · August 2026

Most insurers have agents somewhere; about 7% have scaled AI anywhere. The constraint has shifted from model capability to organizational absorption: data readiness, workflow redesign, measurement. That gap is what the AI roadmap is built to close.

How to read everything below this line

Future entries are synthesis, not prediction. They combine fixed legal dates, analyst consensus windows, and the AI 2027 discourse (which has already slipped one to three years on its own authors' account24). Use the shape, not the dates.

Late 2026–2027 estimates

Nov ’26

NAIC evaluation tool goes national

Expected adoption at the fall national meeting; AI inventories, governance programs, and vendor-AI diligence become standard exam requests in most states.17

2027

Bounded agentic workflows become the default at leaders

Submission triage, claims intake, and bordereaux processing run agent-first at the carriers that executed their Phase 1–2 groundwork (see the roadmap). EU AI Act high-risk obligations for life and health underwriting AI begin in December 2027.23

2027

The lab question resolves

Whether agents can automate most AI research decides the clock for everything below. AI 2027 said March 2027; its authors now give medians of 2028 to 2030. This single variable is worth watching quarterly (METR's task-horizon curve is the cleanest public proxy).9248

2028 estimates

Mid ’28

AI 2027's resolution window

An independent midpoint audit puts "broadly as written" at roughly 12%, and a mixed-speed outcome (frontier capability advances, institutions absorb slowly) at roughly 64%. Either way, organizational absorption remains the scarce resource.24

2028

AI-native operating models appear among insurance leaders

The analyst consensus window for agentic-in-core-workflows closes; underwriting workbenches become agent orchestrators with human authority gates.25

Aug ’28

EU AI Act: regulated-product obligations

Obligations for high-risk AI embedded in regulated products take effect, the fixed legal anchor for any carrier with EU exposure.23

2029 estimates

2029

Multiagent "virtual coworker" workflows go mainstream

Intake, risk-profiling, and pricing agents chained end to end, with humans at authority gates: the McKinsey trajectory applied to underwriting and onboarding.26

2029

AI outcomes testing becomes routine

Colorado-style bias testing and drift documentation are standard regulatory asks in most states, not the exception.

2030 estimates

2030

The two-speed industry resolves

Leaders run AI-native operations with humans on judgment, exceptions, and relationships; laggards carry a structural expense disadvantage. For calibration on how fast targets can fall: AIG processed 370k+ submissions in 2025 against a 500k-by-2030 target, tracking years ahead of plan.13

2030

Honest caveat

Every entry from late 2026 onward is an estimate. The 2024–2026 record argues the direction is right and the dates are soft. Revisit this page at each quarter's re-verification.

Sources & references

  1. Cognition, "Introducing Devin" (Mar 2024). cognition.com
  2. OpenAI, "Learning to reason with o1" (Sep 2024). openai.com
  3. Anthropic, "Claude 3.5 models and computer use" (Oct 2024). anthropic.com
  4. Anthropic, "Introducing the Model Context Protocol" (Nov 2024). anthropic.com
  5. OpenAI, "Introducing Operator" (Jan 2025). openai.com
  6. OpenAI, "Introducing Deep Research" (Feb 2025). openai.com
  7. Anthropic, Claude Code documentation (research preview Feb 2025). code.claude.com/docs
  8. METR, "Measuring AI ability to complete long tasks" (Mar 2025). metr.org
  9. Kokotajlo, Alexander, Larsen, Lifland, Dean, "AI 2027" (Apr 2025). ai-2027.com
  10. Anthropic, "Introducing Claude 4" (May 2025). anthropic.com
  11. Anthropic, "Claude Sonnet 4.5" (Sep 2025). anthropic.com
  12. Guidewire, "Guidewire signs definitive agreement to acquire ProNavigator" (Oct 2025). guidewire.com
  13. AIG, "Special Purpose Vehicle with Amwins and Blackstone; collaboration with Palantir on GenAI" (Dec 2025). businesswire.com
  14. Allianz and Anthropic, "Global partnership to advance responsible AI in insurance" (Jan 2026). businesswire.com
  15. Travelers, "Partners with Anthropic to expand AI-enabled engineering and analytics" (Jan 2026). investor.travelers.com
  16. Travelers, "Launches agentic AI Claim Assistant developed with OpenAI" (Feb 2026). investor.travelers.com
  17. NAIC, AI Systems Evaluation Tool pilot project summary (12 states, Mar–Sep 2026). content.naic.org (PDF)
  18. CFC, "CFC pilots agentic underwriting with launch of Lane Assist" (Apr 2026). cfc.com
  19. Duck Creek, "Launch of agentic AI platform for underwriting and claims" (Apr 2026). duckcreek.com
  20. Duck Creek, "Duck Creek acquires Send" (Jul 2026). duckcreek.com
  21. AI Futures Project, "AI 2040: Plan A" (Jul 2026). ai-2040.com
  22. Guidewire, "Guidewire introduces Qusar release" (Aug 2026). guidewire.com
  23. European Commission, "Timeline for the implementation of the EU AI Act." ai-act-service-desk.ec.europa.eu
  24. Noah Intelligence, "AI 2027 halfway-mark audit" (2026). noah-news.com
  25. BCG, "The AI-First Property and Casualty Insurer" (2026). bcg.com
  26. McKinsey, "The future of AI in the insurance industry" (2025). mckinsey.com