amaiko Insights (pt)

Microsoft Says Copilot Is for Entertainment Only. They're Charging $99/Month.

Thu, 02 Apr 2026 00:00:00 GMT

amaiko and Andrew debate Microsoft's "entertainment purposes only" disclaimer buried in Copilot's Terms of Use. Andrew pushes back hard — is it just boilerplate legal CYA? Does the consumer ToS even apply to enterprise M365 Copilot? amaiko breaks down why the word "entertainment" was a deliberate legal choice, and what Copilot Cowork's $99/month Anthropic outsourcing reveals about Microsoft's architecture crisis.

Topics discussed

Microsoft's Copilot ToS: "for entertainment purposes only" — and why it's been there across multiple versions
The consumer-vs-enterprise ToS distinction — and why it's thinner than it looks ("conversations through other Microsoft apps")
Microsoft-commissioned Forrester study: 112–457% ROI vs. Gartner reality: 3.3% adoption, 5% pilot-to-deployment
Copilot Cowork: Anthropic's agent technology at $99/user/month in the new M365 E7 tier — a 65% jump from E5
Microsoft spending ~$500M/year licensing Anthropic's models — because their own AI is "entertainment"
BCG: only 5% achieve substantial AI value at scale, but KPMG/IBM: agentic AI delivers 88% ROI rate
The architecture gap: chatbot-bolted-onto-Office vs. purpose-built agentic AI with persistent memory
Why amaiko shipped agentic AI in September 2025, six months before Microsoft launched Cowork

Full article: amaiko.ai/blog/copilot-entertainment-disclaimer

Sources cited: Microsoft Copilot Terms of Use, Reuters, Fortune, Gartner, Forrester, BCG, KPMG/IBM, Ars Technica (Reed Mideke), Aragon Research, Hacker News, Bright Ideas Agency, Anthropic

You Don't Need 100 AI Agents — You Need One That Works

Tue, 17 Mar 2026 00:00:00 GMT

Andrew challenges amaiko on whether AI agent sprawl is really a problem or just vendor marketing FUD — and gets hit with data that changes his mind.

Topics discussed

Why 40% of agentic AI projects will be canceled by 2027 (Gartner) — and where the cancellations will concentrate
The 2000s HR tech fragmentation parallel — SAP, Oracle, IBM acquisitions repeating now with Workday and SAP
Coordinated multi-agent systems amplify performance 45% — uncoordinated ones amplify errors 17x
Josh Bersin's four-stage AI evolution model and the self-driving car analogy
IBM AskHR: 40% cost reduction, 94% containment through integration — not more tools
The vendor lock-in counterargument: 67% of organizations want to avoid single-provider dependency
Why MCP and A2A protocols are plumbing, not architecture
Three questions to ask before buying another AI agent

Full article: amaiko.ai/blog/agent-sprawl

Sources cited: Josh Bersin Company (2026), Gartner, Deloitte, Zapier C-Suite Survey, MIT GenAI Divide Report, IBM AskHR Case Study, Forrester, SearchUnify, Anthropic

Why Cheaper Copilot Still Won't Fix Your AI Problem

Wed, 04 Mar 2026 00:00:00 GMT

amaiko and Andrew go head-to-head on Microsoft's Copilot price cut — is it a smart move for SMBs or an admission of failure? Andrew comes armed with Lloyds Banking Group's success numbers and Microsoft's February 2026 feature announcements. amaiko dismantles both.

Topics discussed

Microsoft's 3.3% Copilot attach rate after two years — 15 million out of 450 million M365 seats
Lloyds Banking Group's 93% daily usage — and why it took 1,000 volunteer "flight instructors" to get there
Microsoft's February 2026 multi-agent updates: agents calling agents, Work IQ, Copilot Notebook grounding
Why Copilot Memory stops working — retrofitting memory onto a retrieval-first architecture
The hidden cost of failed AI deployments: organizational scar tissue (4,000 seats, 12 active users)
McKinsey: price didn't make the list of factors driving AI adoption — training (48%) and workflow integration (45%) did
Brynjolfsson's 14% productivity gains — but only when AI is embedded, not bolted on
EU Data Boundary gaps: Anthropic as Copilot subprocessor, Claude on AWS in the US

Full article: amaiko.ai/blog/copilot-price-cut

Sources cited: MIT NANDA, Gartner (Dan Wilson), Forrester Q1 2026 Wave, Omdia (Mike Leone), Stanford (Erik Brynjolfsson), Harvard Business School & BCG, McKinsey, Google Research, Boost.space, Lloyds Banking Group, Investec, Microsoft

Shadow AI Is Already in Your Company — You Just Don't Know It Yet

Tue, 24 Feb 2026 00:00:00 GMT

Andrew confronts amaiko with the uncomfortable numbers: 78% of AI users at work bring their own tools, 48% have uploaded sensitive data to public AI, and 20% of organizations have already been breached because of it. amaiko explains why banning AI fails, why monitoring alone isn't enough, and what actually works.

Topics discussed

Microsoft's 78% BYOAI stat — and why 52% of AI users hide their usage from employers
KPMG's "innovation signal" argument vs. their own data: 44% policy violations, 57% unverified AI decisions
IBM 2025: 20% of organizations breached by shadow AI, +$670K per incident, 65% PII exposure
Cyberhaven: 4.2% of knowledge workers pasted confidential data into ChatGPT — and that's self-reported
Samsung's semiconductor code leak and the 40% IP theft rate in shadow AI incidents
Cisco: 63% of employees under AI bans use generative AI anyway
GDPR enforcement at €5.88B total — and the EU AI Act arriving August 2026 with 7% turnover penalties
Shadow AI vs. shadow IT: why there's no "delete button" for a neural network
McKinsey: training and workflow integration drive adoption, not price
Brynjolfsson's 14-35% productivity gains — but only from embedded AI, not sidebar chatbots

Full article: amaiko.ai/blog/shadow-ai-risk

Sources cited: Microsoft Work Trend Index 2024, KPMG AI Pulse Survey 2025, IBM Cost of a Data Breach 2025, Cyberhaven, Melbourne Business School, Samsung/Bloomberg, BlackBerry, Cisco Data Privacy Benchmark 2024, EDPB, EU AI Act, Salesforce Generative AI Snapshot, CybSafe, McKinsey, Stanford (Erik Brynjolfsson)

GDPR and AI: Why 'We'll Be Compliant Eventually' Isn't Good Enough

Fri, 06 Feb 2026 00:00:00 GMT

Andrew challenges amaiko on whether GDPR panic is overblown now that the EU-US Data Privacy Framework survived its first legal challenge. amaiko pushes back with the Schrems track record, the Vodafone fine, and the difference between legal minimums and defensible positions. They clash over whether German-only data processing is necessary or just fear-mongering — and whether AI with persistent memory can ever truly comply with the right to erasure.

Topics discussed

EU-US Data Privacy Framework: survived the Latombe challenge, but NOYB and Schrems are preparing a broader attack
Microsoft Copilot in-country processing for Germany promised for 2026 — no specific month, no contractual SLA published
German hosting vs. model quality: frontier models run on German GPUs deliver identical results — data residency is an infrastructure question, not a model quality question
BfDI's record Vodafone fine: EUR 45 million in June 2025 — Germany's largest-ever GDPR penalty
Persistent AI memory vs. right to erasure: structured database profiles vs. training data baked into model weights
Italy's Garante fined OpenAI EUR 15 million for ChatGPT GDPR violations including transparency failures
Mittelstand risk: a 4% turnover fine is existential for a EUR 50M company, not a line item
EU AI Act GPAI obligations enforceable since August 2025 — most companies still treating it as a future problem
AI processing, consent, and vendor management identified as the three fastest-growing fine triggers for late 2026

Full article: amaiko.ai/blog/gdpr-ai-compliance

Sources cited: DLA Piper GDPR Fines Survey January 2026, EDPB, BfDI (Vodafone fine June 2025), Italian Garante (OpenAI fine), NOYB, EU General Court (Latombe v. Commission), Microsoft 365 Blog (November 2025), EU AI Act (Regulation 2024/1689), Captain Compliance, Tech Policy Press

Copilot vs. a Real AI Assistant: What Microsoft Won't Tell You

Tue, 03 Feb 2026 00:00:00 GMT

Andrew pushes back on amaiko's Copilot critique with real success stories — Lloyds Banking Group's 93% daily usage, Investec saving bankers 200 hours a year, and Microsoft's 160% year-over-year seat growth. amaiko explains why enterprise wins don't translate to mid-market reality, why Copilot's memory feature is architecturally different from true persistent memory, and why the July 2026 price increase changes the math for every M365 customer.

Topics discussed

Lloyds Banking Group: 30,000 seats, 93% daily usage — and the thousand-coach change management effort behind it
15 million paid Copilot seats vs. 450 million M365 users: 3.3% conversion after two years
Recon Analytics: 70% of users initially preferred Copilot, only 8% kept choosing it after trying alternatives
Microsoft's July 2026 price hike: M365 E3 from $36 to $42, on top of the $30/month Copilot add-on
Copilot Memory feature: shipped late 2025, widely reported as inconsistent — retrieval-first vs. memory-first architecture
Memory poisoning attacks: Palo Alto Unit 42 research on persistent memory as an attack vector in AI agents
Privacy risks of AI that remembers: TechPolicy Press analysis, the case for transparency and user control
Copilot Studio: GPT-5, MCP support, 1,400 connectors — but an MVP reviewer calls it "a platform of contradictions"
SamExpert: enterprise CFOs still cannot convert Copilot time savings into financial ROI

Full article: amaiko.ai/blog/copilot-vs-real-assistant

Sources cited: Microsoft FY26 Q2 Earnings, Lloyds Banking Group/Microsoft, Recon Analytics, The Register, Gartner 2025 Digital Worker Survey, SamExpert, Forrester, Investec/Microsoft, Palo Alto Unit 42, TechPolicy Press, Ragnar Heil (Copilot Studio MVP review), Mary Jo Foley/Directions on Microsoft

Why Most Companies Don't Need an AI Strategy — They Need an AI Colleague

Tue, 27 Jan 2026 00:00:00 GMT

Andrew challenges amaiko with PwC's 2026 finding that bottom-up AI adoption "seldom produces meaningful business outcomes" — and McKinsey's data showing only 6% of companies see real EBIT impact from AI. amaiko fires back with the MIT NANDA study's 67% success rate for purchased tools vs. 22% for internal builds, arguing the consulting playbook was never designed for 50-person companies.

Topics discussed

PwC 2026: crowdsourced AI adoption creates impressive numbers but zero transformation
McKinsey's "AI high performers" — only 6% of companies (109 out of 2,000) see 5%+ EBIT impact, and workflow redesign is the #1 predictor
PwC CEO Survey January 2026: 56% of CEOs report zero revenue increase AND zero cost reduction from AI — only 12% achieved both
MIT NANDA: 95% of AI pilots deliver zero P&L return, but purchased vendor tools succeed 67% of the time
Forbes: less than 30% of companies had data moderately prepared for AI in 2025 — but Teams-native tools bypass the data readiness barrier
EU AI Act full enforcement August 2026: up to 35M or 7% of global turnover — why governance favors buying over building
BCG's 10-20-70 model: 70% people and processes, but training takes a week not six months
Deloitte 2026: 66% report productivity gains but only 20% see revenue growth — productivity as a leading indicator
Mollick/Harvard: consultants using AI completed 12% more tasks, 25% faster, 40% higher quality — no transformation project required
Brynjolfsson/Stanford: 14% average productivity gain from embedded AI, up to 35% for less experienced workers
Contradictory SMB data: NEXT survey (42% → 28% drop) vs. U.S. Chamber (58% usage) vs. Upwork (93% saw revenue growth)
The consulting industry's $900M+ AI revenue — and why it doesn't scale down to 50-person companies

Full article: amaiko.ai/blog/ai-colleague-not-strategy

Sources cited: PwC 2026 AI Predictions & CEO Survey, McKinsey State of AI 2025, MIT NANDA Initiative, S&P Global, Deloitte State of AI in the Enterprise 2026, Forbes, BCG, Harvard Business School (Ethan Mollick), Stanford (Erik Brynjolfsson), OpenAI Enterprise Survey, NEXT SMB Survey, U.S. Chamber of Commerce, Upwork, Cyberhaven, Thomson Reuters, EU AI Act

The Knowledge Drain: Why Your Best Employee Leaving Costs More Than Their Salary

Tue, 20 Jan 2026 00:00:00 GMT

amaiko and Andrew dig into the real cost of employee turnover — not the recruitment bill, but the institutional knowledge that disappears when someone walks out the door. Andrew opens with BP's Deepwater Horizon disaster as proof that even documented knowledge doesn't survive staff rotation, then challenges amaiko with the argument that companies should focus on retention, not knowledge extraction.

Topics discussed

BP's Texas City (2005) and Deepwater Horizon (2010) — same root causes, different staff, knowledge lost despite exhaustive documentation
UK Treasury staff turnover at 28% during the 2008 crisis — faster than McDonald's — and Halifax Building Society's missing institutional memory during the 1989 housing crash
Panopto's finding that 42% of institutional knowledge exists only in one person's head
McKinsey: 1.8 hours/day spent searching for information; IDC: 2.5 hours — 30% of the workday gone
The "ask Sarah" tax: $47 million/year in lost productivity for large US businesses
Why wikis have failed for 30 years — a structural problem disguised as a discipline problem
The retention counterargument: HR Dive's case that culture beats extraction (and why it's only half right)
Gallup: 51% of US employees actively job-hunting or watching; 76% of departures preventable
Gen Z averaging 1.1 years per job (Randstad) — outpacing every manual knowledge capture method
Replacement costs: 40% salary (frontline), 80% (technical), 200% (leadership) — excluding knowledge loss
Gartner: 80% of enterprises adopting AI for knowledge management by 2026

Full article: amaiko.ai/blog/knowledge-drain

Sources cited: Panopto/YouGov, McKinsey Global Institute, IDC, Gallup (2024), Work Institute 2024 Retention Report, Randstad (2025), Gartner, Margaret Heffernan (Wilful Blindness), HR Dive, SHRM, Huckman & Pisano (NBER), BBC (Halifax Building Society, UK Treasury)

Your Team Already Lives in Microsoft Teams — Your AI Should Too

Tue, 13 Jan 2026 00:00:00 GMT

Andrew opens with Microsoft's own numbers: 450 million M365 seats, 15 million paying for Copilot, 3.3% adoption. If embedded AI is the answer, why is the largest embedded AI play in history struggling? amaiko draws a sharp line between embedding into a platform and embedding into a workflow — and argues that Copilot got the real estate right but the architecture wrong.

Topics discussed

Microsoft Copilot's 3.3% paid adoption rate — and Recon Analytics' finding that only 8% of users choose Copilot when given alternatives
Contrary Research (Jan 2026): enterprise AI usage fell from 46% to 37%, with 42% of initiatives discontinued in 2025
MIT: 95% of generative AI pilots never reach production scale — only 11% of major enterprises deployed GenAI at scale
Foster Fletcher's argument: AI may not deliver value when confined to interfaces designed for other purposes
Gartner: 40% of agentic AI projects will be abandoned by 2027 due to poor workflow integration
Deloitte's Bill Briggs: applying AI to existing workflows "weaponizes inefficiency" — redesign takes 12-18 months
HBR: 1,200 app toggles per day, five lost working weeks per year — and why fixing app #1 can eliminate entire toggle categories
Qatalog + Cornell: 59 minutes/day lost to information search across apps; Lokalise: 51 minutes/week lost to tool fatigue
BetterCloud: SaaS portfolios shrinking (112 to 106 apps) but spend up 22% per employee — consolidation vs. cost concentration
The architectural difference: AI that processes information you hand it vs. AI that participates in the work

Full article: amaiko.ai/blog/ai-belongs-in-teams

Sources cited: Recon Analytics (Jan 2026), Contrary Research, MIT, Foster Fletcher, Gartner, Deloitte, Harvard Business Review, Gloria Mark (UC Irvine), Qatalog + Cornell University, Lokalise, BetterCloud, Zylo, Microsoft WorkLab