Agency_

Guides/Case Study
Primary-Source Case Study & Roadmap • 2024–2026

Case Study: Building An Agentic Agency

The full architectural roadmap from The Hobo SEO Quadrilogy, the Skills for LLMs hypothesis, Hobo Web Agentica, Searchable.com, and the March 2026 SMX Paris keynote to the 18 earliest public dispatches from @Hobo_Web defining Agentic-First HITL.

Author: Agency•Last Updated: November 15, 2026•In Guides•RSS Feed
SMX Paris Keynote
March 9, 2026

Unveiled Agency on stage: next year agents will be ubiquitous—the last year without an always-on agent.

Cycle Compression
2 Days → 10 Mins

Deep research, reading, fact-checking, and HCU/Google-leak quality rating unified in one prompt.

Sept 2025 → Marketplace
Skills for LLMs

From the Sept 2025 Skills hypothesis & Agentica to paid, installable practitioner workflows.

Governance Thesis
Agentic-First HITL

Agents operate the tools; the human codifies philosophy, spots hurdles, and manages the agency.

Why Document the Full Roadmap & Earliest Build Log?

Most retrospective product histories rewrite early experiments to look inevitable after the fact. With Agency, every intellectual stepping stone was published in the open by creator and Lead AI Architect Shaun “Hobo” Anderson (@Hobo_Web)—spanning two years of primary-source Google leak forensics, four books, an early modular LLM skills architecture, platform-scale AEO advisory work, a Paris keynote forecast, and 18 timestamped build-in-public dispatches before late-May beta acceleration.

Viewed end to end, the roadmap shows why Agency never resembled a thin chat wrapper around SEO metrics. Long before public launch, its core foundations were forged sequentially: ground-truth ranking forensics (The Hobo Quadrilogy), modular, installable “Skills for LLMs” (Agentica), Answer Engine visibility (Searchable.com), and Agentic-First Human-in-the-Loop (HITL) governance.


Architectural Lineage • 2024–2026

The Roadmap to Agency: From the Quadrilogy & Agentica to SMX Paris

How forensic search research evolved step by step into a self-serve, 81-employee Agentic Operating System.

Stage 01May 2024 – Summer 2025Forensic Ground Truth

U.S. v. Google DOJ Antitrust Trial & Content Warehouse API Leak

Before writing a single line of multi-agent orchestration code, Shaun “Hobo” Anderson spent more than a year auditing sworn testimony from the U.S. Department of Justice antitrust trial alongside the 14,014 leaked Google Content Warehouse API attributes—isolating Q* site quality, P* popularity, NavBoost, Goldmine, explicit siteAuthority, and the LLM-evaluated contentEffort signal.

Stage 02July – October 2025The Hobo Quadrilogy

Codifying the New Reality Across Four Primary-Source Volumes

Between July and October 2025, Anderson published The Hobo SEO Quadrilogy—the four-book diagnostic ruleset that later became the deterministic scoring backbone of Agency’s 38-dial Quality Rater and specialist prompts:

  • Hobo Strategic SEO 2025 (July 2025): Established an evidence-based ranking lexicon built from DOJ trial exhibits and confirmed leaked systems (NavBoost, Goldmine, siteAuthority).
  • Hobo Strategic AI SEO 2025 (August 2025): Coined the Cyborg Technique (Human SEO–AI Symbiosis), defined Third Point Emergence Failure (cascading hallucination drift across multi-step LLM chains), and introduced Creative Inference Optimisation and Disambiguation.
  • Hobo Beginner SEO 2025 (September 2025): Rebuilt foundational search education strictly from sworn Google v. DOJ ground-truth testimony rather than industry folklore.
  • Hobo Technical SEO 2025 (October 2025): A 900+ page forensic reference analyzing the 14,014 leaked Google attributes and popularizing the Content Effort hypothesis.
Stage 03September 2025 OnwardsSkills Hypothesis & Agentica

Hypothesising the “Skills for LLMs” Paper & Launching Hobo Web Agentica

Recognizing that ad-hoc, one-shot chat prompts collapsed under context pollution and Third Point Emergence Failure, Anderson hypothesised a “Skills for LLMs” paper and architecture from September 2025 onwards: packaging human domain mastery into modular, installable capability units—structured multi-stage instructions, verification rubrics, and tool bindings—rather than disposable prompts. That architecture shipped immediately as Hobo Web Agentica: pre-engineered forensic SEO, E-E-A-T, and Content Effort skill modules turning frontier reasoning models (such as Gemini Deep Research) into disciplined specialist auditors.

Stage 04Late 2025 – Early 2026AEO at Platform Scale

Searchable.com Advisory & Answer Engine Optimization

Named a Top AI SEO Expert in 2025 by Primary Position and Searchable.com, Anderson worked alongside Searchable.com (founded by Sam Hogan) to bridge ground-truth Google leak signals—contentEffort, siteAuthority, and Signal Coherence—with commercial Answer Engine Optimization (AEO) and Answer Share tracking across ChatGPT, Claude, and Perplexity. Searchable.com now connects natively inside Agency’s Analyst & AEO Apparatus.

Stage 05January – February 2026Stealth Engineering

Nine-Week Solo Sprint on Codename: Agency & Elsa

Starting in January 2026 in Greenock, Scotland, Anderson began coding Codename: Agency to unify The Quadrilogy’s scoring laws, Agentica’s installable LLM skills, and zero-drift Case Notes ledgers inside one self-serve operating system—placing multimodal AI Company Secretary Elsa between a human CEO and a multi-department AI workforce.

Stage 06March 9, 2026 (2:00 PM CET)SMX Paris Keynote

SMX Paris 2026 Keynote Publicly Introduces Agency

On stage at Search Marketing Expo (SMX) Paris during his keynote “Decoding the Google Leak: What It Means for SEO in 2026,” Anderson unveiled Agency publicly for the first time—warning the audience that within the next year AI agents would be ubiquitous, and that 2026 is the last year you wouldn’t have an always-on agent.

SMX Paris Keynote Forecast (March 9, 2026): Introducing Agency on stage, Shaun Anderson told attendees that within the next twelve months AI agents would become ubiquitous—and that 2026 is the last year any serious practitioner or business would operate without an always-on agent.

Stage 07March 21 – April 30, 2026Build-in-Public Arc

18 Earliest @Hobo_Web Dispatches, thisisagency.ai & Agentic-First HITL

Twelve days after SMX Paris, public field dispatches documented the paid practitioner skills layer (March 21), live 10-minute AEO/SEO and Google-leak quality rating demos (March 24–25), the “Codename: Agency” essays (March 27–30), official registration of thisisagency.ai (March 31), AI Ethics Manager Meg (April 7), the “Silver Era of Vibe Coding” (April 13), and the formal definition of Agentic-First HITL (April 30).

Stage 08May 1 – June 10, 2026 & BeyondBeta & Full Launch

Private Beta, Zero-Sycophancy Self-Audit & The 81-Agent OS

Private beta opened May 1, 2026, followed by limited public beta (May 26–29, including Agency’s unvarnished 14/100 baseline Q* self-audit) and full public launch on June 10, 2026—scaling into an 81-employee Glass-Box Agentic Operating System (70 Office specialists + 11 Call Centre agents) operated under Candlestickoperations Ltd.


Phase I • March 21, 2026

1. The Installable Skills Layer & Closed Weekend Alpha

The earliest public dispatch in this series (March 21, 2026) did not lead with model benchmarks—it led with incentives and codified human expertise. Users would be able to install themselves, their proprietary workflows, and any external API directly into Agency as reusable skills that other operators could install. Unlike open giveaway ecosystems that strip specialists of commercial leverage, the premise was explicit: “you get paid for your experience.”

Four hours later (status/2035416072606634458), Anderson alluded to a deeper architectural layer—known to roughly 15 private testers entering a weekend evaluation—that went beyond a conventional multi-agent digital swarm: coordinated specialist hierarchy, persistent state, and an executive control plane rather than unguided agent chatter.

Phase II • March 24–25, 2026

2. Ten-Minute Proof of Work: AEO, Organic Links & Google-Leak Quality Rating

Three days later, theoretical claims turned into live recorded demonstrations:

  • AEO via SEO in One Prompt (Demo 1): Demonstrated end-to-end Answer Engine Optimization (AEO) and SEO research-to-publication in roughly ten minutes from a single executive instruction, compressing what previously demanded one to two full days of manual deep-research prompting, reading, fact-checking, and editing (follow-up). Crucially, Agency “knows what it is fixing.”
  • Real-World Link Earning (March 24): Within hours of shipping output from the workflow, Agency-produced research earned its first organic editorial citation—validating the system as an authority-driven link builder rather than a thin-content mill.
  • Helpful Content Update & Google-Leak Simulation (Demo 2 / 3): On March 25, Agency analyzed a domain hit by Google's 2023 Helpful Content Update, scoring it against attributes surfaced in the Google Content Warehouse leak to emulate quality rating in ten minutes—the direct forerunner to today's Website Quality Rater Emulator.
Phase III • March 27–31, 2026

3. “Codename: Agency” & Mapping the Cyborg Exoskeleton

Between March 27 and March 31, the architecture was formalized in two flagship essays on the Hobo Web site: “Codename: Agency – A Weapon of Mass Disruption?” (March 27) and “Codename: Agency – Mapping The Cyborg Exoskeleton for the Agentic Age” (March 30).

That same weekend, Anderson highlighted a newly circulated Google multi-agent research paper (status/2037666484550304005), noting that its hierarchical specialist topology described almost word-for-word the multi-agent digital marketing agency already running in production tests. On March 31 (status/2038971556802515252), he distilled the entire user-experience philosophy into a single sentence: augment the human to expand their agency and autonomy while stripping out the inefficiencies of operating software tools.

Phase IV • April 12–30, 2026

4. From “HITL Errors” to Agentic-First HITL

April's dispatches shifted focus from what the agents could do to what the human must become when managing an autonomous workforce:

  • Accountability Remains Human: A wry April 12 note on a “HITL error” (status/2043260765021356221) and an April 22 reflection (status/2046980039455977945) drew a sharp boundary: Agency can act as a patient teacher when no mentor is available, yet understanding the strategic hurdles ahead remains the unmistakable duty of the Human-in-the-Loop.
  • Interoperable Skilled Agents: By April 26 (status/2048482243572560324), the roadmap extended to a networked HITL “cyborg apparatus” in which skilled agents are employable via API and can be introduced to one another to collaborate across organizational boundaries.
  • The April 30 Definition of Agentic-First HITL: Closing the foundational spring arc (status/2049899432280244706), Anderson defined Agentic-First HITL in terms that still govern Elsa and the 81-agent office today:
“In Agentic-First HITL, the human no longer uses tools directly; the agent does. The human trains the agent, codifies their philosophies, and manages an Agency rather than prompting. I wish you good fortune in the clone wars to come.”
— Shaun Anderson (@Hobo_Web), April 30, 2026

Primary-Source Archive: Earliest @Hobo_Web Posts (March–April 2026)

Every publicly archived dispatch below links directly to the original timestamped post on X. Volume and granular feature documentation accelerated sharply from late May 2026 onward as the public beta opened.

#01Skills & Stealth Alpha
View on X (709555) →

The Installable Skills Layer & Practitioner Economics

Key early statement on the skills layer: users will be able to install themselves and their skills (including any API) into Agency so other people can install them too. Contrasts this with pure open systems by noting that “you get paid for your experience.”

#02Skills & Stealth Alpha
View on X (634458) →

Closed ~15-Person Weekend Alpha Beyond a Digital Swarm

Mentions a bigger, still-secret aspect of Agency (known only to ~15 people at the time) that goes beyond the capabilities of a large multi-agent digital swarm; notes those testers would be able to comment once they tested it that weekend.

#04Live 10-Min Demos
View on X (307223) →

Compressing Two Days of Research & Fact-Checking Into Ten Minutes

Follow-up noting that the single agentic SEO auditing process (previously a day or two of deep research prompts, reading, fact-checking, and editing) now takes about ten minutes—and that Agency “knows what it is fixing.”

#07Live 10-Min Demos
View on X (803609) →

Demo 3: Helpful Content Update & Google Leak Quality Rating in 10 Minutes

Demo 3 output: SEO analysis of a site hit by the 2023 Helpful Content Update, using attributes and scores from the Google leak to simulate quality rating, completed in ten minutes.

#18Agentic-First HITL
View on X (244706) →

Defining “Agentic-First HITL” — Managing an Agency Instead of Prompting

Defines “Agentic-First HITL”: the human no longer uses tools directly; the agent does. The human trains the agent, codifies their philosophies, and manages an Agency rather than prompting. Ends with: “I wish you good fortune in the clone wars to come.”


From Late-May Public Beta to the 81-Agent Operating System

These 18 spring dispatches established the blueprint before public beta velocity took off in late May 2026. Every concept sketched in March and April now ships as production infrastructure inside Agency:

Shaun Anderson

Shaun Anderson

Shaun Anderson, AKA @hobo_web, is the Lead AI Architect behind Agency (thisisagency.ai). Shaun has over 25 years of experience in forensic SEO auditing. He was a primary investigator of the Google Search Leak of 2024, the writer of The Hobo SEO Quadrilogy, and a Keynote speaker at SMX Paris 2026.

Disclosure: I use generative AI when specifically writing about my own experiences, ideas, stories, concepts, tools, tool documentation or research, across the Hobo Web and Agency platforms. My tool of choice for this process is Agency—the publishing architecture I created and licensed to Candlestickoperations Ltd—using Google Gemini Pro (2.5 / 3) Deep Research and Reasoning and later models (as default). I have been writing about Google Search for 2 decades. This content is a direct result of cutting-edge HITL AI content production, using a process I designed myself and described in my 2025 ebook, Strategic AI SEO. Also note: This content has not been created to “pass off” as human, or “write as Shaun Anderson” (me). It is raw, factually correct content. Agency can produce content with certain styles, but it is not my aim to fool anyone with it. This is the state of AI content in 2026, an example of the default Agency creative output. All content was conceived, directed, edited, fact-checked and verified as correct by me (and is under constant development). This article is amongst the first “beta-test” of Agency, the autonomous AI-powered Agency as presented in my March 2026 SMX Paris keynote presentation. Edited by Shaun Anderson AKA Hobo. Corrections welcome. Read my AI policy.

Sign Up for Agency Updates

Subscribe for product updates, new AI workflows, and SEO research. Unsubscribe anytime.

Agency (Thisisagency.ai) is in public beta. Agency is an AI workplace simulation. Agency was conceived, bootstrapped, and architected solo by Shaun Anderson AKA Hobo (@hobo_web). “This Is Agency” and “Agency” are trading names of Candlestickoperations Ltd. Registered in England and Wales, Company No. 17503419. Registered Office: 3rd Floor, 86-90 Paul Street, London, England, EC2A 4NE.
All team members are AI entities powered by Google Gemini technology (by default). When API are available, Agency is a multi-provider, multi-model, multi-modal, multi agent system.