The State of Play

AI email drafting and triage is a practice caught in an unusual bind: the technology is production-ready everywhere, yet confident organizational adoption remains elusive. Every major email platform — Gmail, Outlook, Apple Mail — ships AI composition and prioritisation features to billions of users, and 40% of business professionals report using AI for email drafting weekly. The tooling works. The problem is everything around it.

Field research reveals a productivity paradox: AI-suggested replies increase email volume by 38%, but 61% of those replies contain no new information, and teams using them experience 22% longer resolution times. Speed gains mask effectiveness losses. Meanwhile, 55% of consumers view AI-generated emails negatively, and documented professional liability cases — a $285k retainer lost to AI-drafted contract errors, a $220k penalty for misrepresentation — give compliance teams pause. Security compounds the hesitation, with 88% of organisations reporting AI-powered phishing attacks that erode trust in automated email systems.

This keeps the practice bleeding-edge despite mature vendor capabilities. Individual knowledge workers gain real time savings, but the barriers to organisational scaling are non-technical and resistant to further platform investment.

Platform feature maturity reached a new phase in July 2026. Google's Help Me Write advanced to GA custom refinement (July 14-20 rollout), enabling free-form editing prompts ('add deadline,' 'make warmer') replacing preset buttons; this capability is now available across consumer (AI Plus/Pro/Ultra) and enterprise (Business/Enterprise/Frontline) tiers. Google simultaneously monetized email management by introducing AI Inbox as a tiered premium feature (Plus $19/month through Ultra $60+/month), positioning inbox prioritization as a core consumer-facing AI value proposition. Microsoft's Copilot Chat in Outlook expanded from single-thread reasoning to full inbox, calendar, and enterprise data reasoning across all platforms (Android, Windows, iOS, Mac, Web) in early July, while the company's new Copilot Notebooks capability (mid-July GA) enables Outlook emails as grounding sources for synthesis, presentations, and briefing generation—email data now feeds directly into multi-modal AI output workflows. These July 2026 updates signal ecosystem-wide feature maturity: all three major platforms (Google, Microsoft, Apple) now ship production-grade email triage, drafting, and synthesis capabilities integrated directly into primary email clients.

Measured deployments demonstrate that organisation-scale automation is technically feasible. A Fortune 500 financial services company deployed autonomous email agents (40K+ employees in March 2024, scaled to 12K across three regions by July 2026) achieving 42% email volume reduction, 92% accuracy, 22-minute response time (from 8.7 hours prior), and 38% job satisfaction increase with $180K ROI in the 90-day pilot. Rachio (1M+ smart-home users) deployed Crescendo.ai's multi-channel triage agents with 95-99.8% accuracy and 30% cost reduction, enabling a single customer support leader to manage full volume. These deployments show that machine-learning triage operates successfully at scale when governance is prioritized.

However, adoption barriers crystallize even as capabilities mature. Only <4.5% of Microsoft 365's 450M customers pay for Copilot features; of that sliver, only 1% are weekly-active users despite three years of GA, ubiquitous feature integration, and vendor promotion. This adoption failure—97% customer rejection despite captive user base—signals product-market fit failure at the organisational level despite cutting-edge capabilities. Threat landscape escalation compounds hesitation: July 2026 threat intelligence documents ConsentFix attacks hijacking M365 accounts in three seconds via dragging a callback link into browser (OAuth token exposure), EvilTokens phishing with encrypted malware reveal, and the first fully agentic ransomware attack where an AI agent adapted and succeeded 31 seconds after an initial failed login attempt. These email-specific attack vectors undermine organizational confidence in email-AI deployments. Critical barriers from earlier assessments remain unresolved: authenticity concerns (USC Marshall research: supervisor sincerity drops from 83% to 40-52% with heavy AI use), deliverability paradoxes (79% of B2B professionals report deliverability collapse despite 70% using AI), trust collapse (Deloitte: 89% confidence drop in agentic systems), and security governance gaps (92% of security professionals concerned about AI agents lacking guardrails).

An emerging best-practice pattern contradicts the platform-vendor default: triage-first (not drafting-first) maximizes ROI. Production architects document a classify-label-draft safety framework where AI reads and prioritizes messages but humans approve before sending. Practitioners' measured evidence shows dramatic ROI: a marketing consultant's one-week inbox audit (587 emails: 68% noise, 21% need reply, 11% need decision) proved that triage-first system redesign cut daily inbox time from 2h40m to 50 minutes—ROI concentrated in filtering and routing, not in writing speed. McKinsey data (28% of workweek on email) supports this: the leverage is in reducing volume that needs attention, not in accelerating replies. Despite this emerging consensus among practitioners, platform vendors continue shipping drafting-first features (Help Me Write refinement, Copilot Chat summarization), and organizational adoption metrics show 30% write with AI but only 22% read with AI—a reversal of where the ROI actually sits. Measurement infrastructure for email AI adoption remains immature: organizations lack confidence-threshold frameworks, audit logging of autonomous sends, and ROI attribution models that distinguish triage ROI (volume reduction, routing accuracy) from drafting ROI (speed). These barriers—adoption product-market failure, threat landscape escalation, authenticity/deliverability trade-offs, triage-vs-drafting misalignment, governance/measurement gaps—explain why platform feature maturity has not translated to organizational scaling despite measured technical success in pilot deployments.

• 2022-H2: Microsoft and Google made suggested replies and text predictions generally available in Outlook and Gmail respectively, establishing email drafting AI as a standard platform feature; however, narrow capability scope and limited personalization kept adoption passive and use cases restricted to quick, templated responses.
• 2023-H1: Google expanded Gmail's capabilities with "Help me write" for full-compose assistance; Mailbutler released Smart Assistant with smart compose and voice-aware rewriting; Superhuman demonstrated advanced deployments with context-aware drafting; research documented 6.7 billion Smart Reply sends by 2017 and explored LLM-mediated approaches, signalling growing vendor investment and platform maturation beyond template replies.
• 2023-H2: Research and deployment data revealed dual signals: Clemson peer-reviewed study confirmed productivity gains (reduced workload, improved task performance), and Nature research documented billions of daily Smart Reply uses in Gmail; simultaneously, real-world deployments exposed quality and interpersonal barriers—cold-email case studies showed failed attempts due to lack of authenticity, Microsoft Outlook's suggested replies feature shipped with bugs in production, and user research indicated AI-generated emails triggered lower cooperation ratings if detected. Platform expansion continued (Outlook rolling out to Office Insiders) but adoption remained cautious.
• 2024-Q1: Superhuman launched Instant Reply with 2x faster email writing and full-draft generation (not templates), indicating vendor acceleration in context-aware drafting. Outlook's suggested replies expanded to broader Outlook.com population, revealing user adoption friction and feature confusion. Critical assessment emerged documenting limitations of AI email tools—lack of emotional depth, generic responses, and poor handling of complex contexts—identifying barriers that constrained broader deployment despite vendor investment and platform availability.
• 2024-Q2: Platform consolidation accelerated: Google rolled out Gemini AI in Gmail (web and mobile) with summarization and contextual smart replies (May–June); Microsoft shipped Copilot in Outlook with 30% time-reduction claims (May); Apple signalled incoming Mail enhancements via Siri and Smart Replies (WWDC rumors, June). Third-party vendors expanded (MailMaestro reported 300,000+ teams, 8+ hours saved weekly). Industry-wide adoption signals emerged (45% of marketing teams using AI in email, up from 31% planning investment). However, critical assessments continued identifying limitations: originality concerns, skill-development risks, ethical challenges, and limited contextual nuance constraining broader rollout. The ecosystem showed vendor investment momentum across all major platforms, but quality and user-perception barriers persisted.
• 2024-Q3: Google achieved general availability of contextual Smart Replies powered by Gemini (September 26), and Microsoft launched new Outlook for Windows with integrated Copilot (August 1), signalling ecosystem-wide platform maturation. Simultaneously, enterprise adoption faced headwinds: 60% of IT leaders had started Copilot pilots but only 6% were planning large-scale deployments due to data security, governance, and ROI concerns; a pharmaceutical company CIO cancelled Copilot for 500 employees after six months due to cost ($30/user/month) and perceived low value. Consumer sentiment remained negative (55% viewed AI-generated emails unfavorably), citing authenticity concerns (41%), and Google rolled back a quick-reply feature for Workspace customers mid-rollout due to performance and quality issues. The practice exhibited classic bleeding-edge tension: vendor momentum and capability maturation competing against persistent adoption barriers, cost uncertainty, and quality concerns preventing confident enterprise scaling.
• 2024-Q4: Platform-native email AI features continued steady availability: Google's support documentation confirmed Smart Compose and Smart Reply GA across Gmail, while Microsoft maintained Suggested Replies in Outlook; third-party vendors (Mailbutler, MailMaestro) reported ongoing feature development. Adoption metrics showed measured but cautious growth: 15% of U.S. internet users reported using AI to write emails (October survey), with 50% explicitly uninterested and 25% considering adoption. Critical assessments from domain experts intensified: Chicago Booth Review published analysis documenting that AI-generated emails produce clichéd, fact-checking-poor content unsuitable for professional communication; industry research (Validity study, 47% of marketers using AI) addressed delivery concerns, confirming well-formed AI emails reach inboxes successfully. By year-end 2024, the practice remained in bleeding-edge tier characterized by entrenched vendor feature sets, modest consumer adoption rates, measurable technical maturity in deliverability, and persistent quality/authenticity concerns cited by both practitioners and end-user sentiment research. Organizational deployment barriers (cost-value uncertainty, data governance) prevented confident tier advancement despite stable platform feature availability.
• 2025-Q1: Google launched contextual Smart Replies to Gmail Workspace Business and Enterprise tiers (March 2025), integrating Gemini AI natively without separate Gemini add-on—marking ecosystem consolidation. Third-party vendors reported concrete deployment outcomes: Revo reached multiple organizations with 60% inbox time reduction and 4-hour daily savings; Valentin Solutions case study documented 23% revenue uplift and 98% follow-up rates via AI triage implementation. However, critical assessments accelerated: The Spinoff published opinion warning against AI email drafting despite platform availability, citing authenticity and relationship concerns; futurecoworker.ai's enterprise analysis documented persistent barriers (data governance, broken-process amplification, ROI uncertainty). Platform-scale adoption remained constrained: 50% of U.S. users uninterested in AI email composition; enterprise pilots faced barriers (only 6% planning large-scale Copilot rollouts despite 60% having started). Practice remained in bleeding-edge: vendors delivered mature, production-ready features across all major platforms, yet organizational and user adoption barriers prevented confident scaling despite demonstrated technical capability.
• 2025-Q2: Platform vendor feature evolution continued: Google announced Contextual Smart Replies GA in April 2025 and personalized smart replies at Google I/O (May) with context pulled from past emails and Google Drive; Mailbutler achieved GA status for Smart Assistant across three major email platforms (Apple Mail, Outlook, Gmail) with GPT-4o; Outlook refined text prediction. Third-party vendors sustained niche success (Revo, MailMaestro, Valentin). However, fundamental adoption barriers remained unresolved: Alexander von Humboldt Institute research (April 2025) documented deskilling, tonal dissonance, and emotional detachment as persistent adoption barriers; organizational adoption slowed with 69% of companies reducing AI investment pace (May 2025 survey). Enterprise readiness remained cautious despite 43% of IT leaders (survey of 4,009) expecting AI-driven email automation to dominate in five years. Practice remained in bleeding-edge tier: ecosystem-wide platform consolidation and feature sophistication had reached maturity, yet authenticity concerns, skill erosion fears, and organizational ROI uncertainty prevented confident enterprise scaling.
• 2025-Q3: Platform feature expansion completed: Microsoft introduced Copilot for Service with email summarization and contextual drafting (July 2025); research documentation of real-world productivity outcomes showed ~30 fewer minutes weekly on email management from Copilot deployments. However, critical adoption barriers crystallized: ASU research documented hidden risks (trust erosion, perceived manipulation, reputation costs, error propagation) in AI email drafting at scale; organizational adoption remained stalled with persistent data governance and ROI concerns. Platform ecosystem maturity reached saturation point with all major vendors (Google, Microsoft, Apple) and third-party tools (Mailbutler, Revo, Superhuman) offering GA production-ready capabilities, yet enterprise scaling remained constrained by non-technical barriers to authentic communication. Practice remained in bleeding-edge tier: technical capability fully mature across all channels, but adoption barriers non-technical and resistant to platform investment.
• 2025-Q4: Platform vendor feature hardening continued: Microsoft confirmed GA of Copilot in Outlook with real-time email refinement and automated meeting triage (December 2025). However, field research exposed critical limitations undermining adoption: study of 47 professionals found AI-suggested replies increased daily volume 38% but 61% of those replies contained zero new information; teams using Smart Reply experienced 22% longer email resolution times, revealing a productivity paradox where speed gains masked effectiveness losses. Critical practitioner assessments intensified: HR director documented AI drafting generating legally inaccurate accommodations guidance and authentication concerns. Quantitative triage benchmarks showed efficiency upside (6-10x speedup, 95-98% accuracy) but only in controlled vendor-promotional contexts without independent validation at enterprise scale. By year-end 2025, the practice exhibited mature platform consolidation (all major vendors GA) offset by evidence of unintended consequences—shallow replies, degraded decision-making, quality risks in sensitive contexts—preventing confident tier advancement. Bleeding-edge tier status reflected maximum technical capability with minimum confidence in organizational deployment impact.
• 2026-Jan: Platform ecosystem consolidation reached completion: Google launched major Gemini integration into Gmail (January 8) with Help Me Write, AI Overviews, Suggested Replies, and Proofread rolling to 3 billion users; Microsoft maintained Copilot in Outlook GA; vertical-market specialization emerged with Triage launching for insurance agencies (January) claiming 50% email handling reduction and $151,667 annual per-agent savings. Broader adoption metrics showed sustained growth with quality caveats: 87% of businesses use AI in email workflows (Knak survey) but only 6% are high performers; named case studies (Amazon 95% reduction, Google Cloud 90% reduction, FTI 316x capacity) documented efficiency in marketing contexts. Gallup survey of 22,000 U.S. workers confirmed 12% daily AI use and one-quarter several-times-weekly adoption. Adoption barriers remained intractable: authenticity concerns unresolved (55% negative consumer sentiment, 41% authenticity concerns), productivity paradoxes from late-2025 research continued (higher volume, lower information content, longer resolution times), and marketer concerns emerged about inbox AI mass-unsubscribe features. Practice remained bleeding-edge: all major platforms shipped mature GA features but authenticity, cost-value uncertainty, and productivity paradoxes prevented confident enterprise scaling.
• 2026-Feb: Platform maturity continued with Google Research documenting Smart Reply at 12% of all Gmail Inbox replies (February 10), signaling sustained adoption; Mailbutler released product updates with composing status indicators (February 24); industry analysis concluded "Email AI is no longer optional" within major platforms. Adoption metrics accelerated: 40% of business users employed AI for email drafting weekly; 25%+ of inboxes used AI for auto-summarization/categorization; AI saved professionals half a day weekly on email tasks. However, critical barriers surfaced: 88% of organizations experienced AI-powered phishing/deepfake attacks undermining trust (Osterman Research); real-world failure cases documented professional liability—marketing agency lost $285k retainer, architecture firm faced $220k penalty due to AI-drafted errors. Practice remained bleeding-edge with mature technical consolidation offset by emerging security, authenticity, and professional liability concerns challenging organizational deployment confidence.
• 2026-Mar: Enterprise deployment signals accelerated but infrastructure brittleness emerged as critical risk. Microsoft Copilot reached 60% of Fortune 500 companies (up from 35% prior year) with 20-40% productivity gains validated in pilots; email drafting explicitly cited as validated high-value use case. Abnormal AI's specialized graymail triage demonstrated maturation with behavioral AI and measurement infrastructure (12% inbox volume reduction). McKinsey 2026 deployment testing validated 45-minute daily triage savings and 80% draft accuracy. Google's automatic email summarization (March 27) rolled to production but carried documented accuracy limitations. However, critical infrastructure failures exposed system vulnerabilities: Gmail spam-filter collapse (January 24) disabled email categorization for 1.8 billion users; Outlook post-security-update freezing disrupted mail access. These cascading failures revealed brittleness in cloud-dependent AI email systems and dependency on fragile authentication infrastructure. Enterprise adoption benchmarking positioned email drafting as table-stakes copilot capability but identified fundamental limitation: ROI attribution difficult because underlying workflows unchanged (only 27% process-level adoption, 4% transformational). Professional liability and security concerns remained acute barriers. Practice remained bleeding-edge: platform vendor capabilities mature and enterprise-scale deployments underway, yet infrastructure reliability, quality assurance on automatic features, ROI attribution challenges, and persistent trust/liability concerns prevented confident organizational scaling.
• 2026-Apr: Adoption metrics confirmed broad but trust-challenged deployment: 82% of professionals use AI email features with 38% of all email traffic now AI-generated, yet a Quinnipiac poll found trust fell to 31% with 68% verifying AI outputs before use — a 31-point usage-trust gap. Google Cloud Next 2026 confirmed Workspace Intelligence (Help Me Write, thread summarization, cross-tool draft generation) rolling to production across Gmail and Docs, while AI-powered thread summarization is now standard across all three major email platforms (Gmail, Outlook, Apple Mail). A peer-reviewed 1,100-person survey documented a new adoption friction: trust in managerial emails drops from 83% (low-AI teams) to 40-52% (high-AI teams), and while most recipients cannot detect AI-drafted messages, disclosure triggers ratings of "lazy" and "insincere" — a disclosure penalty that creates a structural dilemma for professional deployment. Named production deployments continued: Senetic automated responses across 2M+ annual emails via RAG; supervised deployment guidance shows 60-70% response-time reduction but identifies legal and conflict contexts as unsuitable for automation. The core pattern held: tooling works at scale but organizational confidence remains undermined by trust deficits, verification overhead, and the emerging reputational cost of disclosed AI authorship.
• 2026-May: Agentic email automation reached production at scale. Microsoft deployed agentic Copilot for Outlook (April 27) with inbox triage, draft composition, rule creation, and calendar conflict resolution—advancing from single-turn assistance to persistent background automation. Copilot user base reached 20+ million paid seats with 250% YoY growth and weekly engagement exceeding Outlook benchmarks; Accenture's 743K-employee deployment (largest to date) reported 97% completing routine tasks faster. Independent ROI measurement emerged: Fusion Computing's Q1 2026 study of 11 SMB deployments showed 2-4x Year 1 net ROI with email triage accounting for 45% time reduction and 90-day payback in 9 of 11 customers. Google's Workspace Intelligence GA (May 18) unified Gmail, Drive, and Calendar into semantic layer enabling context-aware drafting; Microsoft released Copilot for Sales with integrated email intelligence and Outlook summarization (May 21). However, two critical barriers crystallized. First, security: PromptArmor discovered zero-click prompt injection in Superhuman Mail's triage summarization allowing exfiltration of sensitive emails from 40+ messages via hidden form submission. Proofpoint's 1,400-respondent security survey found 87% of organizations deployed AI assistants yet only 42% experienced confirmed incidents, with 63% of those incidents involving email—indicating widespread deployment ahead of security readiness. Phishing attacks have matured: AI-generated phishing emails now defeat traditional detection signals, creating new triage challenges. Second, authenticity and quality: Macquarie peer-reviewed study (N=547) documented "AI penalty" effect—communication perceived as 19% less trustworthy and less useful for knowledge uptake—with perverse "disclosure paradox" where recipients penalize disclosed AI despite 74% believing disclosure should occur. MIT NANDA research found 95% of enterprise AI pilots return zero measurable productivity gain; field deployments reveal quality drag where AI-drafted emails increase volume but 61% contain no new information. A real-world sales copilot deployment failed entirely due to hallucinations, bias, and tone issues. Email deliverability is being rewritten by AI triage: generic openers now 'deliverability liabilities' because Gemini summarization changes content visibility. Gartner projects 40% of agentic AI projects to be cancelled by end 2027 due to escalating costs and inadequate controls. Market adoption data shows 68% of enterprise teams using AI email features with 40-60% time reductions, yet organizational confidence remains constrained by trust deficits and quality assurance gaps. Practice remained bleeding-edge: infrastructure maturation and broad tool adoption alongside newly discovered security vulnerabilities, quality paradoxes, and persistent organizational trust deficits resisting confident enterprise scaling.
• 2026-Jun: Platform feature expansion and measurement gap exposed. Microsoft Copilot's May redesign (task-aware workspace on email canvas) drove +30% usage growth in Outlook, the second-highest lift across Microsoft 365 apps; 20M+ paid Copilot seats with 90%+ Fortune 500 adoption confirmed. Google expanded Workspace Intelligence (June 3) to consolidate Gmail as selectable data source in Ask Gemini, signalling full ecosystem-level triage maturity with cross-app context awareness. Fusion Computing's 40-person financial planning firm case study (60-day deployment) documented 15-20 hours/week recovered via governance-first Copilot rollout, adding named evidence of SMB-scale ROI. Productivity measurement confirmed 43% of Copilot users apply the tool to email summarization, with 65% reporting time savings; however, Gartner data revealed only 3% rate adoption as 'significant value,' with a 7-licensing-trap analysis identifying measurement failures and governance delays as the primary cause of ROI shortfall. Folderly research (June 18) documented a sharp deliverability paradox: 70% of B2B professionals use AI for email, yet 79% report deliverability collapse, with 75% of heavy AI users saying outbound email is 'getting harder' — tool adoption itself creating friction. USC Marshall academic research confirmed the trustworthiness crisis: AI-drafted emails significantly damage sender trustworthiness perceptions, with peer research documenting supervisor sincerity ratings dropping from 83% (low AI) to 40–52% (high AI). Gartner projects AI tools will handle 30% of enterprise email by EOY 2026 — a supply-side signal — while trust collapse (Deloitte: 89% confidence drop in agentic systems) and security constraints (92% of security professionals concerned about AI agents) continue to cap organizational deployment confidence.
• 2026-Jul: Platform feature maturity reached iterative editing stage. Google released Help me write iterative capabilities (June 25) enabling specific edits via prompt bar without full regeneration; 900M Gemini MAU (May 2026) and 120K+ enterprise adopters document platform-scale deployment. However, field evidence exposed critical maturity ceiling: ActivTrak behavioral study (443M hours, 1,111 companies) showed AI adoption correlates with email volume +104% and focused work sessions -9%—productivity paradox where speed gains are masked by attention fragmentation and increased task density. Adoption failure crystallized around two core barriers. First, market saturation: personalization inflation collapse documented 2018 (5–10 cold emails/week per VP) to 2026 (60–120/week); buyers developed 'AI blindness' filter with response rates below 3% industry-wide; email provider enforcement (Gmail/Yahoo engagement scoring) now penalises high-volume senders. Second, organizational rejection: only 3.3% of 450M M365 users converted to paid Copilot despite captive user base; 97% conversion rejection signals product-market fit failure despite platform advantage. On-premise deployment alternatives emerged (Mail Sentinel: 15,000+ classified messages, confidence thresholds, no cloud exposure) addressing security constraints, while agentic triage governance matured (four-tier framework with confidence thresholds, approval_rules, escalation logic). Critical security vulnerability persisted: CVE-2026-26133 (prompt injection in email summarization) patched March 2026 but architectural vulnerability remains unresolved. Email-specific adoption remained shallow: 30% write with AI, 22% read with AI (despite 88% broad AI use), with only 25% 'AI Superworkers' showing workflow redesign; integrated tools showed 83% productivity vs standalone 20%—integration gap identified as primary limiting factor. Practice remained bleeding-edge: platform capability matured to iterative editing and agentic governance, yet market saturation, organizational adoption failure (3.3% conversion), unresolved security architecture, attention fragmentation paradox, and shallow workflow integration prevented tier advancement. Mid-to-late July updates deepened both monetization and integration: Google introduced tiered AI Inbox pricing ($19-$60+/month) alongside GA custom refinement for Help Me Write (free-form edit prompts, July 14-20 rollout), while Microsoft expanded Copilot Chat in Outlook to full inbox/calendar/enterprise-data reasoning and launched Copilot Notebooks using Outlook emails as grounding sources for briefings. Case studies reinforced measured triage ROI: a Fortune 500 financial-services deployment (40K+ employees) cut email volume 42% with 92% accuracy and $180K pilot ROI, and Crescendo.ai's triage agents at Rachio hit 95-99.8% accuracy with 30% cost reduction. Threat intelligence darkened the security picture — ConsentFix account hijacking, EvilTokens OAuth theft, and the first fully agentic ransomware attack — sustaining the trust barrier to organizational adoption.
• 2026-Aug: Platform buildout continued: Google's July 15 Gmail AI deployment (Help Me Write, AI Inbox, Suggested Replies, Gemini Spark agent) reached full-stack GA, and Superhuman shipped an MCP server enabling Claude/ChatGPT to read, draft, and send mail directly, signalling a shift from standalone tools to AI-assistant-native email workflows. Trust and saturation evidence hardened in parallel: cold-email reply rates fell from 0.50% to 0.35% over 2025 (a "tragedy of the commons" from AI-drafted outreach), a security researcher disclosed a since-patched ChatGPT Workspace Agents phishing flaw, and adoption surveys confirmed the confidence gap — 64-65% of professionals use AI in email workflows but only 1.1% trust it for autonomous sending, with named deployments (A.Team on Superhuman Mail: 4.4 hrs/week saved; Copilot 365: 353% 3-yr ROI but only 20-35% regular license usage) showing real but narrow gains.