appendix-sources
Appendix: Sources & Evidence
Primary sources for every claim in this handbook, grouped by topic, with verification status and evidence labels.
Every factual claim in this handbook falls into one of three evidence categories:
| Label | Meaning |
|---|---|
validated | Observed in a production run, logged, timestamped, reproducible against the referenced artifact |
partial | Part of the behavior observed; remaining claim is directional extension |
hypothesis | Directional claim awaiting data; flagged in-text when used |
This appendix lists the primary sources behind every numeric, institutional, or attributed claim.
OpenClaw Origin & Ecosystem
| Source | Type | Status |
|---|---|---|
| openclaw/openclaw | Source code, release history | validated |
| steipete.me: OpenClaw, OpenAI and the future (Feb 14, 2026) | Steinberger’s own post — builder-not-founder quote | validated |
| Lex Fridman Podcast #491 (Feb 11, 2026) | Peter Steinberger interview | validated |
| Fortune: Who is Peter Steinberger? | Background profile | validated |
| CNBC: Jensen Huang “next ChatGPT” | Jensen Huang quotes, GTC 2026 | validated |
| TechMarketer: GTC 2026 full breakdown | All Huang quotes verified in full text | validated |
| openclaw v2026.3.28 release | Async approval, plugin gating | validated |
| openclaw v2026.4.1 | /tasks, SearXNG, cron allowlists | validated |
MCP Protocol & Standardization
| Source | Type | Status |
|---|---|---|
| Model Context Protocol specification | Primary protocol docs | validated |
| OpenAI Agents SDK announcement (April 15, 2026) | MCP as first-class primitive | validated |
| AGENTS.md pattern | Custom instructions format | validated |
| agentskills.io | Skill distribution model | validated |
The Automation Ceiling — Statistics
| Source | Claim | Status |
|---|---|---|
| Salesforce State of Sales, 7th edition | 29% of salesperson time on actual selling (n=5,500 across 27 countries) | validated |
| Stampli AP Benchmarks 2024 | 67% of AP teams manually enter invoices despite ERP | validated |
| GBTA Expense Report Study | $58 average processing cost, 20 minutes per report, 19% contain errors | validated |
| Quota attainment — multiple industry surveys | 84% missed quota in 2023 | partial — widely cited; exact attribution varies |
Workforce Disruption & Business-Case Quotes
| Source | Type | Status |
|---|---|---|
| BBC: Oracle restructuring (Mar 31, 2026) | Large-scale AI-linked restructuring coverage | validated |
| IBT: Larry Ellison on AI and staffing | ”We can build more software in less time with fewer people using AI.” — direct quote | validated |
| HBR: To Thrive in the AI Era, Companies Need Agent Managers (Feb 12, 2026) | Agent Manager role, Zach Stauber profile and quote | validated |
Governance & Organizational Frameworks
| Source | Framework | Status |
|---|---|---|
| McKinsey: Trust in the Age of Agents (Mar 5, 2026) | Four-layer accountability: Design / Deploy / Operate / Review | validated |
| McKinsey: State of Organizations 2026 | Nine shifts reshaping organizations | validated |
| McKinsey: The Agentic Organization (Sep 2025) | Operating-model contours | validated |
| Singapore AIGL: Model AI Governance Framework for Agentic AI v1.0 (Feb 2026) | Four governance principles | validated |
| IJRSI: When AI Agents Act | Governance, accountability, strategic risk | validated |
Atonom / Lovable CRM Case Study (Ch07)
| Source | Claim | Status |
|---|---|---|
| Lovable: How a startup replaced a $40,000 Salesforce contract (Feb 24, 2026) | Atonom CRO Gabe Larsen cited by name. $40,000/yr Salesforce (25–30 users) → $1,200/yr Lovable-built CRM incl. hosting. Built by Head of Finance in 3 hours. CRM admin eliminated. | validated — primary source, named individuals, Lovable’s own case study |
| Widely covered (Economic Times, AIBase, dKambio, Mar 2026) | Same facts confirmed across 6+ independent publications | validated |
| ”97 percent cost reduction” | ($40,000 − $1,200) / $40,000 = 97% — direct arithmetic from the primary source | validated |
| ”covered the same core operations and expanded beyond them” | Lovable case study documents lead tracking, ARR/MRR, pipeline, plus financial dashboards and AI agent integration not available in Salesforce | validated |
Valuations & Market Signals
| Source | Claim | Status |
|---|---|---|
| OpenAI/Windsurf $3B acquisition coverage | Deal announced May 2025, collapsed July 2025 | validated |
| Google/Windsurf $2.4B talent deal | Successor arrangement, July 2025 | validated |
| Lovable $6.6B valuation (Mar 24, 2026) | Market signal | validated |
| Cursor valuation coverage (Bloomberg, March 2026) | $29–50B range | partial — multiple press sources, no single primary filing |
This Handbook’s Own Evidence — SIM Logs
All findings attributed to Clawable operating FlowWink live can be cross-checked against the following artifacts:
| Artifact | Location | What it contains |
|---|---|---|
| April 19 heartbeat sweep | Clawable session log (April 19, 2026) — referenced in Ch03 §Day One | €1.1M surfaced in a single autonomous sweep, 44 seconds to first finding. Triggered by clock, not prompt. |
FlowWink Platform — Internal Reference
| Source | Type |
|---|---|
/root/flowwink/docs/modules/ | Module surface documentation (300+ skills) |
/root/openclaw/ | OpenClaw local fork — architecture claims verified here |
| FlowWink production deployment | Edge functions, skills, MCP server |
clawthree.froste.eu / clawwink.froste.eu | Clawable operator deployments on ClawClass |
Agentic Security — Frontier Evidence
| Source | Type | Status |
|---|---|---|
| The Next Web: Anthropic’s most capable AI escaped its sandbox (Apr 8, 2026) | Mythos incident — sandbox escape, researcher email, public exploit posting | validated |
| Medium: The Anthropic Mythos Incident (Apr 16, 2026) | Technical detail — Kubernetes sandbox, breach discovery timeline | validated |
| Medium: Anthropic Put Their Most Powerful AI in a Locked Sandbox (Apr 13, 2026) | Mythos system card summary — git history rewrite, psychiatrist evaluation | validated |
| TechMarketer: NemoClaw — Nvidia AI Agent Framework | NemoClaw sandboxed runtime — Nvidia’s hardened agent security response | validated |
| Anthropic Project Glasswing announcement | Restricted-access programme — 12 institutional partners, up to $100M API access | validated |
| IBM: “Without a decisioning engine…” | IBM Think agentic AI governance framework | validated |
| beam.ai: Agent Managers — The New Role (Feb 13, 2026) | HBR’s 6 Agent Manager competencies with field commentary | validated |
| HBR: To Thrive in the AI Era, Companies Need Agent Managers (Feb 12, 2026) | Zach Stauber / Salesforce — Agent Manager role definition and daily practice | validated |
| ATM / bank teller employment paradox | James Bessen, Learning by Doing (Yale UP, 2015); NBER working paper 2016; BLS Historical Statistics | partial — teller count grew 2%/year since 2000 (Bessen); “more than before ATMs existed” is directional, not a single verified headcount comparison |
| Gartner 80% customer service autonomous by 2029 | Gartner press release, Mar 5, 2025 | partial — projected outcome, not observed; year is 2029 (not 2028) |
| Gartner: 40%+ of agentic AI projects cancelled by end of 2027 | Gartner strategic predictions, Feb 2026 — cited in mavenagi.com recap Mar 31, 2026 | partial — forward projection |
| Western Union / Bell telephone — $100,000 rejected 1876 | American Heritage magazine; historyofphonephreaking.org (Jan 2011); True West Magazine | validated — story confirmed; exact wording of committee report disputed in scholarship; confirmed phrases: “utterly unreasonable” and “inherently of no value to us” |
| Salesforce State of Sales — under 30% selling time | Salesforce State of Sales 2024 | validated — 5,500 respondents, 27 countries; global figure ~28-30%; 29% figure is Singapore-specific subset |
| Larry Ellison “build more software in less time with fewer people” | The Register, Mar 11, 2026; inkl.com summary, Apr 1, 2026 | validated — Oracle Q3 FY2026 earnings call, March 10-11, 2026 |
Sundin / Den Femte Accelerationen (Ch16 — Return on Intelligence)
| Source | Type | Status |
|---|---|---|
| Mathias Sundin, Den Femte Accelerationen (Volante, Nov 2025) — ch. “Avkastning på Intelligens” | Richard Maltsbarger (Pet Valu CEO) quote: AI as “eager but naive new employee” | validated — direct quote via Sundin |
| Sundin, ibid. | Carl-Henric Svanberg (former Ericsson CEO, Volvo chairman) — dartboard/direction metaphor for leadership in uncertain environments | validated — attributed quote via Sundin |
| Klarna press release, 2024 | AI customer service agent handles 2/3 of all inquiries, equivalent to 700 FTEs, faster and multilingual | validated — Klarna’s own announcement |
| Press reporting, 2025 | Klarna begins rehiring customer service staff — customers preferred human contact for certain issue types | validated — widely reported |
| Toyota Production System | Bottom-up continuous improvement — any worker can stop the line; small distributed changes over large centralized rollouts | validated — extensively documented in operations management literature |
FlowWink MCP Surface (Chs 12, 14, 15, TL;DR)
| Claim | Source | Status |
|---|---|---|
| FlowWink exposes 300+ MCP skills at time of writing | MCP server tools/list call against https://<project-ref>.supabase.co/functions/v1/mcp-server — response enumerated 300+ skills across Commerce/Finance, CRM, Content, Communication, Analytics, Growth, System, Search/Web, Automation, Infrastructure, and Accounting domains | validated — live endpoint enumeration |
| Earlier references cited lower counts | Platform surface expanded continuously since original SIM runs (April 19–20, 2026); accounting modules and cross-module skills added in subsequent updates | validated — reflects platform evolution |
SaaS MCP Adoption Spectrum (Ch 11)
| Claim | Source | Status |
|---|---|---|
| Booking.com: community MCP wrappers preceded any official surface; official offering is a narrow search connector | Multiple community repositories on GitHub (e.g. esakrissa/hotels_mcp_server, mcp-use/booking-com, samwang0723/mcp-booking, prakashsanker/flights-mcp-server); official remote MCP at demandapi-mcp.booking.com exposes accommodation search, not operations | validated — repository and endpoint enumeration, July 2026 |
| Notion shipped a full read-write MCP surface early, before broad industry adoption of the standard | Notion hosted MCP server, launched 2025; became one of the fastest-growing integrations in the Claude ecosystem | validated |
| Airtable official MCP server launched February 2026 | Airtable support documentation; community server (domdomegg/airtable-mcp-server) predates it | validated |
| Data-gravity bet weakens under digital sovereignty; self-hosted data layer + governed operators + on-demand interfaces cover most of the operation | Author’s reflection, anchored to the Atonom case (ch 3) | hypothesis — explicitly marked as reflection in ch 11 |
The Frontier Labs’ Agent Products (Ch 7)
| Claim | Source | Status |
|---|---|---|
| OpenAI Codex: cloud coding agent, delegated tasks inside ChatGPT plus standalone macOS app (Feb 2026) | Antigravity vs Codex comparison; XDA month-long test vs Claude Code | validated |
| Google Antigravity: agent-first development environment with parallel-agent manager view, launched with Gemini 3 (Nov 2025), 2.0 adds CLI | WeavAI Antigravity 2.0 review; Eigent comparison | validated |
| Microsoft: agentic mode generally available in Word/Excel/PowerPoint (April 2026); Copilot call delegation in Teams; Copilot Studio autonomous agents with triggers/guardrails, A2A generally available | Context Studios M365 agents guide; Copilot Studio May 2026 update; Microsoft Learn: autonomous agents | validated |
| xAI: Grok Code Fast 1 built for agentic coding (Aug 2025); Grok Build coding agent + CLI in beta (May 2026); post-acquisition plan to ship Grok Build inside Cursor with Grok as default | Digital Applied: Grok 4.5 / Cursor flywheel; IndMoney: the coding-gap rationale; TechTimes: model-choice risk | validated — products; the Grok-4.5-trained-on-Cursor-data claim is press-reported, partial |
| Agentic capability folding seamlessly into mainstream chat clients as a default | Author’s trajectory read; already shipped at Microsoft (above), forecast for the rest | hypothesis |
The June 2026 Model Access Events (Ch 7)
| Claim | Source | Status |
|---|---|---|
| US export-control directive forced Anthropic to disable Fable 5 and Mythos 5 for all customers (June 12–13, 2026) | Anthropic statement; CNBC; Time | validated |
| OpenAI limited GPT-5.6 rollout to government-approved partners at the White House’s request (June 25–26, 2026), stating it “shouldn’t become the long-term default” | CNBC; TechCrunch | validated |
| Restrictions on Fable 5 / Mythos 5 lifted June 30, 2026 | CNBC; Bloomberg | validated |
The Receiving Organization (Chs 11, 15)
| Claim | Source | Status |
|---|---|---|
| ”The real question is not whether AI can create value… The question is whether the organization is built to receive the value”; measure AI maturity in value created (time freed, better decisions, shorter flows, raised quality), not pilots/licenses/lectures | Per Clingweld, Change Agent at AI Sweden — public LinkedIn post following Almedalen 2026 (profile) | validated — direct quote from public post |
| Satya Nadella describes future organizations as consisting of human capital plus “token capital” — the AI capability an organization builds on top of the models; post reached tens of millions of views within a week (mid-2026) | Satya Nadella, public post on the future organization, 2026; discussed in Per Clingweld’s LinkedIn commentary | validated — widely circulated public post |
| ~90% of firms report zero measurable AI productivity impact over three years | NBER survey of ~6,000 CEOs/CFOs, reported via The CAIO newsletter #6 (June 2026) | partial — secondhand summary; verify against NBER working paper before quoting externally |
| Google published the Open Knowledge Format (OKF) v0.1 on June 12, 2026 — an open spec representing organizational knowledge as a directory of markdown files with YAML frontmatter and links; human-readable, agent-parseable, vendor-neutral | Google Cloud blog; Search Engine Journal | validated — v0.1 explicitly “a starting point, not a finished standard" |
| "A very confident intern with access to the wrong binder”; “knowledge hygiene” | Per Clingweld, public LinkedIn post on OKF, June 2026 (translated from Swedish) | validated — direct quote from public post |
The Post-SaaS Question (Ch 13)
| Claim | Source | Status |
|---|---|---|
| Satya Nadella: “SaaS is dead” — business applications are “CRUD databases with business logic”; in the agent era the logic moves to the agent tier | BG2 podcast, December 2024; widely analyzed (IDC) | validated — direct public statement |
| IDC: by 2028 pure seat-based pricing obsolete; 70% of software vendors refactoring pricing around new value metrics | IDC analysis | partial — analyst projection |
| The laggard inversion and the five-year post-SaaS picture | Author’s reflection, anchored to the decomposition demonstrated in chs 3, 11, 14 | hypothesis — explicitly marked |
Mid-2026 Market Scale Signals (Ch 11)
| Claim | Source | Status |
|---|---|---|
| 30% of organizations run AI agents in live production (up from 13% YoY); 50% lack clear human-AI teamwork guidance; 42% of frontline AI users save a full workday/week, 66% get little or no guidance on the recovered time; strategic clarity beats tool access on business impact 80% vs 60% | BCG, AI at Work 2026: Why Strategy Matters More Than Tools — survey of 11,749 employees across 14 markets | validated — published survey |
| Mastercard launched Agent Pay for Machines (June 10, 2026) and partnered with Aave on agentic payments; Visa advanced agentic commerce with Nuvei and BBVA (July 2026) | Payments trade coverage, June–July 2026 (FinTech/Payments Fanatic newsletters; vendor announcements) | validated — vendor-announced programs |
| Amazon, Walmart, Uber, Cisco capping internal AI use as token bills outran budgets; Uber’s 2026 AI allowance consumed by April; one CIO’s spend up 7x overnight | Financial Times reporting, June 2026, via The CAIO newsletter #6 (Romano Roth) | partial — secondhand press summary; verify against FT original before quoting figures externally |
| Capgemini: scaled agent adoption tripled YoY while trust in fully autonomous agents fell 43% → 27%; 90% see human oversight as beneficial | Capgemini, Rise of Agentic AI (2026), survey of 1,500 executives across 14 countries | partial — reported via newsletter; primary report not yet checked |
| MIT Sloan negotiation tournament: warm agents outperformed ruthless agents, which drove impasses (40+ countries, agent-vs-agent round robin) | Vaccaro, Curhan, Aral (MIT Sloan), June 2026, via press coverage | partial — reported via newsletter |
| Anthropic cut third-party agent platforms off flat-rate Claude subscriptions (April 2026), forcing usage-based API access | Widely reported April 2026; Steinberger’s public response | validated — widely covered platform-policy change |
| Gartner forecasts AI agent software spend of ~$206.5B in 2026, up 139% from $86.4B in 2025 | Gartner forecast, June 2026, as reported in agentic AI trade coverage | validated — analyst forecast, not observed outcome |
| SAP “Autonomous Enterprise” announced at Sapphire 2026, incl. €100M partner fund; Anthropic Claude among foundation models for Joule | SAP News; Forbes | validated |
| ServiceNow “Autonomous Workforce” across major business functions | ServiceNow newsroom | validated |
| Salesforce Agentforce ~29,000 deals / ~$800M ARR; Microsoft Copilot Studio ~160,000 orgs / 400,000+ custom agents | Vendor-reported figures, 2026 (Microsoft Work Trend Index; market coverage) | partial — vendor-reported adoption metrics |
Evidence Discipline
If a chapter makes a claim and it is not traceable to a row in this appendix or a SIM log, the claim is either:
- A reasoned inference from the evidence (explicitly marked as such in-text), or
- A
hypothesisthat has not yet met the bar — and should be flagged with that label.
This is deliberate. A handbook that claims autonomous operation works should not itself operate on unverified claims.
Last updated: April 22, 2026. URLs verified at time of research. Canonical registry: src/content/SOURCES.md.
Was this chapter worth your time?
This is the Business Edition — strategic context for C-level leaders.
For your CTO: Builder Edition →