Clicky

Codename: Agency – Mapping The Cyborg Exoskeleton for the Agentic Age

by

The Centaur and the Swarm: How Google’s “Agentic AI” Paper Validates the “Agency” Blueprint.

NOTE FROM SHAUN: Normally, I read Google papers to find out how Google works. It’s strange to read a Google-affiliated academic collaboration that hypothesises about a future with the very systems you’ve actually built and beta-tested in public over 6 months. You’re about to read an article created by the system described as theory in the latest Google paper, one that you will be able to employ in days, not years.

In the history of technology, there are rare moments of inflexion when high-level academic theory and practical engineering arrive at the same conclusion from opposite directions.

I just witnessed one.

On March 19, 2026, researchers from Google and academia published a landmark paper in Science titled “Agentic AI and the next intelligence explosion.” It argues that the “Singularity”—the idea of a single, godlike AI mind—is a myth. Instead, they hypothesise that true superintelligence will be plural, social, and bureaucratic.

While the academics were peer-reviewing this theory, I was busy building it, in public.

For the last year to six months, I’ve been publicly constructing “Codename: Agency” – a multi-agent marketing swarm that doesn’t just “chat” but functions as a digital corporation.

The release of the Google paper serves as a stunning academic receipt, validating everything from the “Marketing Cyborg” philosophy to the granular architecture of its autonomous employees.

Here is how the academic theory maps perfectly to the practical reality of Codename: Agency

This table was made by Claude after reviewing the article and the code base:

Paper Claim Article Claim Agency Code Reality Accuracy
Societies of thought Agents debate strategy Sequential layered verification ⚠️ 80% — “debate” overstates parallelism
Centaur / human sign-off Human CEO final approval Hard-coded approval gate ✅ 100% accurate
Institutional alignment / Meg Meg blocks content, zero-tolerance Meg runs pre-flight, mid-run, post-run ✅ 100% — actually understated
Real-world grounding Leak signals prevent hallucination contentEffort/scoring/Ethan ✅ 95% accurate

That’s almost a complete match, even at this early stage.

“The article is a highly accurate representation. The only honest quibble is “agents debate” which implies parallel deliberation – the reality is more like a sequentially-gated review chain. That’s still a “society of thought” – just more like a law firm than a parliamentary debate.” Claude Opus 4.6

That’s by design, by the way. And Agency does debate, except with hierarchies in place.

I asked @Grok, Claude and Gemini (with access to the actual Agency code) the implications, and this is the hybrid output:

1. The Theory: “Societies of Thought”

  • The Paper: The authors note that advanced models (like DeepSeek-R1) don’t just “think harder”; they simulate internal debates. The paper predicts the rise of “recursive agent ecologies” where AI agents spawn sub-agents to solve complex problems through “collective deliberation.”

  • The Reality (Agentica & The Swarm, 2023–2025): Anderson identified this need years ago. His 2023 work on “Skills” and the 2025 launch of Agentica were built on the premise that a single LLM prompt is insufficient for professional work. He didn’t want a chatbot; he wanted a specialist.

  • The Convergence: In Anderson’s March 2026 demos, we see this “society of thought” live. When “Elsa” (the Agency Secretary) receives a task, she doesn’t just answer. She routes it. Agents debate strategy. A specialist executes, and another verifies. It is the “recursive descent” described in the paper, materialised as a digital office.

2. The Theory: The “Centaur”

  • The Paper: The paper discards the idea of humans being replaced by AI. Instead, it introduces the “Centaur” – a hybrid actor where human intent is amplified by a collective agency of AI systems.

  • The Reality (The Marketing Cyborg, 2025): Throughout 2025, Anderson evangelised the “Marketing Cyborg” technique: the human as the CEO/Strategist and the AI as the “exoskeleton.” (Editor’s note: I made that Marketing Cyborg image for my post on the rise of the marketing cyborg at Searchable.com, but they didn’t want it. It is a little too Sci-Fi for the a-suite, but that IS what is happening.)

  • The Convergence: “Codename: Agency” is not fully autonomous by design; it is a Cyborg Apparatus. In the demos, the human CEO provides the vision and the final sign-off (“The Vision”), while the AI swarm handles the execution. The paper argues this is the only way to scale intelligence safely; Anderson proved it is the only way to scale SEO effectively.

3. The Theory: “Institutional Alignment”

  • The Paper: How do you control a swarm of a billion agents? The paper suggests we need “digital institutions”—structures like courts and bureaucracies that enforce rules, asserting that “Power must check power.”

  • The Reality (Meg and the Ethics Office, 2026): This is perhaps the most striking predictive feature of Anderson’s work. He realised that a swarm moving at hyper-speed produces “slop” if unchecked.

  • The Convergence: Enter “Meg”, the Agency’s Ethics Manager. Just as the paper calls for “institutional alignment,” Anderson built a literal Compliance Department into his software. Meg doesn’t produce content; she blocks it. She checks for bias, missing claims, and E-E-A-T failures before the human ever sees the draft. The “digital courtroom” Google hypothesised is already in session inside Anderson’s dashboard.

4. The Foundation: Grounding in “Physics,” Not Probability

The Google paper argues that these agentic societies must be entangled with the real world. For Anderson, that “real world” was the 2024 Google Content Warehouse API Leak.

While others treated the leak as news, Anderson treated it as training data.

  • 2025 Forensic Work: He spent a year mapping signals like contentEffort, siteFocusScore, and the “Goldmine” system.

  • The Application: When Agency runs an audit in the “10-minute demos” of March 2026, it isn’t guessing. It is simulating Google’s own Quality Raters using the very metrics (attributes/scores) Anderson decoded from the leak.

  • The Result: This provides the “empirical grounding” the academic paper insists is necessary to prevent hallucination.

The Verdict: The Explosion is Here

The Science paper ends with a prediction: the next intelligence explosion will be “plural, social, and deeply entangled with us.”

Shaun Anderson’s timeline – from the static checklists of 2022 to the chaotic, productive, self-governing swarm of Codename: Agency in 2026 – proves they are right. But it also proves that you don’t need to wait for a research lab to build it.

As of March 2026, the “Centaur” (or “Marketing Cyborg”) is no longer a myth. It’s a user role. And the “Society of Minds” is just a login away.

Note from Shaun: You may notice I use a lot of SCI-FI references in my social media and blog posts. I actually like myths and legends too, but the idea of merging humans with myth doesn’t, in my mind, automatically, instantly describe what is happening here. The cyborg technique framework does this in one word, despite the often negative consequences of cyborgs in popular film and art concepts.

Get your cyborgs correct – the Terminator isn’t a Cyborg, Robocop is.

So is the Borg…

Timeline of Convergence

Pre-2025: Early Foundations

2025: Google Leak Forensic Work, Agentica, Marketing Cyborg, AEO and Searchable.com

  • 28 April 2025 – Publishes “Optimise for the Synthetic Content Data Layer Opportunity Gap“, the first attempt at describing in public the abstractive process Anderson was using between 2023-25. The first attempt to describe the process of creative inference optimisation. (Editor note – This was a key moment. The first true understanding of the creative publishing possibilities AI-assisted work afforded. The technique itself was used to create all the Hobo SEO Dashboard documentation in the run-up to April 2025, but this was the first creative work of any substantial depth.)
  • May 2025– Begins probing the Google Content Warehouse 2024 more deeply.
  • 5 June 2025 – Publishes the Automony Gauntlet, a test of the creation and publishing hypotheses laid out in the article “Optimise for the Synthetic Content Data Layer Opportunity Gap“. The first work of pure creative inference published using AI in this manner.
  • June 6 2025 – Continues to deeply focus the latest Google Gemini AI to help decode the Google Content Warehouse data leak attributes, mapping them to signals of E-E-A-T and Google Quality Rater guidelines.
  • July 3 2025 – Publishes E-E-A-T tool in Google Sheets.
  • 4 July 2025 – Publishes “AI Overviews – The Mentions Economy (AEO)”. A piece describing the imminent explosion of mention pollution, e.g., “listicles”.
  • July 10, 2025: Shaun Anderson publishes Hobo Strategic SEO 2025 (free ~194-page ebook, later updated to V2 on October 12, 2025) on hobo-web.co.uk. This is the first major public release where he treats the 2024 Google Content Warehouse API leak not as conspiracy fodder or gossip, but as real “physics” and training data for understanding Google’s ranking system. Key focus: Early forensic mapping of leaked signals such as contentEffort (Google’s LLM-estimated human effort in content), modular ranking components, E-E-A-T as an emergent property of many granular attributes (not a single score), siteFocusScore, title rewriting systems (“Goldmine”), and cross-references with Quality Rater Guidelines + DOJ trial testimony. The Glass Box Moment (The True Starting Point?) He introduces the core concept of “Glass Box” thinking: moving away from the old “Black Box” era of guessing and black-hat heuristics toward treating Google’s algorithm as an understandable, mappable engineering system. The book positions SEO as engineering for a known system rather than speculation, and it becomes the foundational “training material” and philosophical bedrock for everything that follows in 2025–2026. This release kicks off his ambitious Hobo SEO Quadrilogy 2025 (four major free ebooks released throughout the year) and marks the moment Anderson shifts from traditional SEO expert to forensic systems thinker. It directly enables the entire later chain: Marketing Cyborg philosophy (August), Agentica skills (September), Claude Skills leverage (October), Searchable.com advisory role (November), and ultimately the full governed multi-agent Codename: Agency swarm in March 2026. In essence, July 10, 2025, is the single most important inflexion point in the timeline – the day he publicly committed to “Glass Box” forensics and began treating the leak as the new canon of truth. Everything else in 2025–2026 builds directly on this foundation.
  • August 5 2025: Releases Hobo Strategic AI SEO 2025. Anderson publishes Hobo Strategic AI SEO 2025 (a free ~180-page ebook and the second book in the Hobo SEO Quadrilogy). This is the moment the “Marketing Cyborg Technique” moves from private philosophy to a fully public framework. The book formally introduces the core idea: the human remains the strategist/CEO providing vision, intent, taste, empathy, and final sign-off, while the AI swarm functions as a literal exoskeleton — an integrated, always-on layer of specialist agents that dramatically amplifies speed, scale, and signal accuracy without replacing the human.
    Key innovations in this release include:

    • Explicit framing of the human + AI relationship as RoboCop-style cyborg (biological core with cybernetic augmentation and closed-loop feedback) rather than Terminator-style android replacement.
    • Strong emphasis on treating the 2024 Google Content Warehouse leak as “physics,” not gossip: mapping verifiable signals such as contentEffort, siteFocusScore, and E-E-A-T as an emergent property of hundreds of granular attributes to prevent generic “slop” and ensure outputs remain grounded and high-quality.
    • The first public blueprint for what would evolve into Codename: Agency — a governed multi-agent system where the human CEO sets direction and the AI handles recursive execution, internal debate, specialist routing, and compliance. This August 5 release serves as the critical philosophical and technical bridge between the “Glass Box” forensics of July 10 and the full multi-agent swarm of March 2026. It is the exact point where Anderson moves beyond talking about AI as isolated tools and begins describing a self-checking digital corporation living inside the dashboard. By late September 2025, the first pre-packaged “skills” (leading directly into the Agentica launch on September 25) are already being tested publicly, turning the cyborg concept from metaphor into a working, practical prototype. Everything that follows in late 2025 and early 2026 builds directly on this foundational “Marketing Cyborg” release.
  • September 14 2025, Anderson publishes Hobo: Beginner SEO 2025 – the entry-level book in the Hobo SEO Quadrilogy 2025. This free ebook serves as a modern, evidence-based starting point for newcomers, drawing heavily on Google v. DOJ trial testimony and shifting focus toward “people-first” principles rather than outdated myths. It acts as the accessible on-ramp to the more advanced concepts introduced in the Strategic and AI-focused books.

  • September 22 2025, at 14:18, Anderson publishes the very first public Agentica skill: “Rate My Page Quality” — a detailed, 12-criterion prompt framework designed to let LLMs perform a deep, structured audit of any URL. This prompt algorithmically estimates contentEffort, evaluates E-E-A-T signals, and simulates Google Quality Rater thinking using insights drawn from the 2024 Google Content Warehouse leak, official Quality Rater Guidelines, and related sources.

  • September 25, 2025: Launches Hobo Web Agentica (pre-packaged “skills”/multi-stage prompts for turning LLMs into specialised SEO analysts). Publishes the conceptual paper “Agentica: Invoking ‘Skills’ for LLMs and AI Assistants”. Three days later, on 25 September 2025, he officially launches Hobo Web Agentica — a free suite of pre-packaged “skills” (specialised, multi-stage prompt frameworks) — and publishes the conceptual paper “Agentica: Invoking ‘Skills’ for LLMs and AI Assistants”.

    Key innovations in this release include:

    • Shifting from generic single-prompt chatbots to structured, repeatable agent-like workflows that deliver consistent, professional-grade SEO analysis (starting with page-level quality rating).
    • Putting the Marketing Cyborg Technique into immediate practice: the human strategist/CEO stays in full control of vision and final decisions, while the AI skills act as reliable exoskeleton components for execution, research, verification, and grounding.
    • Early demonstration of recursive and specialist behaviour — the prompt breaks down complex quality assessment into clear criteria, cross-references leak-derived signals, and produces verifiable outputs rather than generic slop.
      This September 2025 release (starting with the 22nd) marks the first tangible, publicly available prototype of the multi-agent system that would later become the full Codename: Agency. The “Rate My Page Quality” skill quickly becomes a foundational building block, heavily used in later work and directly feeding into the governed swarm demos in March 2026.

    October 16 2025 – Anthropic launches official Claude Skills. This comes 3.5 weeks after Shaun had already publicly shipped his first Agentica skill on 22 September 2025 at 14:18 and the full Hobo Web Agentica framework plus the conceptual paper “Agentica: Invoking ‘Skills’ for LLMs and AI Assistants” on 25 September 2025. Anderson had been developing and evangelising this structured “skills” approach as part of the Marketing Cyborg Technique since early September.

  • October 20 2025: Releases Hobo Technical SEO 2025 – On 20 October 2025, Anderson publishes Hobo Technical SEO 2025 — the final, massive 900+ page instalment in the Hobo SEO Quadrilogy. This expert-level ebook serves as the comprehensive capstone, delivering an exhaustive forensic analysis of the 2024 Google Content Warehouse leak (including the full set of 14,000+ attributes), modular ranking systems, and advanced technical signals. It completes the journey from beginner-friendly foundations (September) through strategic and AI-focused volumes to full technical mastery. Just four days after Anthropic’s official launch of Claude Skills on 16 October 2025, Anderson notes the announcement and continues to build on the Agentica framework he had already launched in September. He positions Claude’s move as further validation of the structured “skills” approach he had been publicly developing and evangelising for months as a core part of the Marketing Cyborg Technique.

    Key innovations and activities in October include:

    • Deep integration of leak-derived signals (such as contentEffort, siteFocusScore, and the “Goldmine” system) into practical technical SEO workflows using his custom Agentica skills for repeatable, high-precision analysis.
    • Public showcasing of how the Agentica skills enable recursive, specialist agent-like behaviour that produces consistent, high-quality output while minimising slop through verifiable grounding.
    • Additional forensic articles and tools that reinforce the cyborg philosophy: the AI swarm/exoskeleton (powered by custom Agentica skills) handles heavy execution and analysis, while the human CEO provides direction, taste, empathy, and ethical guardrails.
      This October release solidifies the philosophical and technical foundation for the full multi-agent Codename: Agency swarm. It marks the completion of the ambitious 2025 Quadrilogy project and directly enables the Searchable.com advisory role in November–December 2025 and the early 2026 cyborg apparatus developments that follow.

  • November–December 2025: Becomes Special Advisor to Searchable.com, steering E-E-A-T and verifiable Google quality signal grounding (drawing directly from leak analysis). Frames Searchable as the “Cyborg Apparatus” realisation. Notes that “the Agentic Age started for real in December 2025” (tied to Google Antigravity). Recognised as Top Global AI SEO Expert 2025; Hobo blog #1 UK SEO Blog by Feedspot.

  • Throughout 2025: Anderson conducts intensive public forensic analysis of the Google Helpful Content System and Google HCU and the 2024 Google Content Warehouse API leak (2,500+ pages), mapping signals like contentEffort (LLM-estimated human effort in content), modular ranking systems, E-E-A-T as an emergent property of granular signals, title tag rewriting (“Goldmine” system), link building implications, and HCU (Helpful Content Update) patterns. He cross-references with Quality Rater Guidelines, DOJ trial testimony, and real site audits. This becomes foundational “training material” for his later agentic systems and is heavily baked into Agency.

Early 2026: Cyborg Apparatus and SMX Paris

  • January 9, 2026: Publishes “What is Searchable.com? It’s the ‘Cyborg’ Apparatus I Predicted”, linking Agentica, leak work, and cyborg framework.

  • January 15, 2026: Announced as keynote speaker at SMX Paris 2026 — session: “Decoding the Google Leak: What It Means for SEO in 2026” (delivered March 9, 2026).

  • January 19, 2026: Publishes “Shaun Anderson SEO Expert: Work in Public (2025–2027)”, documenting leak analysis as foundation for agentic tools.

  • March 6, 2026 (pre-SMX): Posts on X about “vibe coding” ad agencies and teases Agency free trial launch after SMX Paris session.

  • March 9, 2026: Delivers SMX Paris keynote on Google leak implications for the AI/agentic era (E-E-A-T, contentEffort, verifiable signals in multi-agent systems). Presents the first public screenshots of Agency.

  • March 10–12, 2026: X posts teasing the “full power” of Agency (including “dark (legal) side of SEO” demonstrations of speed/scale) and SMX Paris reflections.

  • March 18, 2026: Publishes on Hobo SEO blog “Codename: Agency – By Your Command”, documenting leak analysis as a foundation for agentic tools.

March 19: Convergence and Public X Demos of Codename: Agency

  • March 19, 2026: Academic paper “Agentic AI and the next intelligence explosion” published in Science.

    • Key direct quotes (aligning with Anderson’s demos and work):

      • Singularity myth: “a single, titanic mind… is almost certainly wrong… will be plural, social, and deeply entangled with its forebears (us!).”

      • Society of thought: Models spontaneously generate “internal debates among distinct cognitive perspectives.”

      • Centaurs/hybrids & explosion already here: Through “orchestration of societies of AI agents… in new ‘centaur’ configurations” and “recursive agent ecologies.”

      • Governance: “Power must check power” via institutional alignment.

  • March 21, 2026: arXiv preprint available.

  • March 23, 2026: Publishes “Meet Elsa: Agency Secretary” (AI receptionist / central nervous system for task routing and structure in the multi-agent swarm). X post announces it.

  • March 24–27, 2026 (peak demo period on X): Anderson undertakes and shares multiple public demos of Codename: Agency on @Hobo_Web, often with screenshots, video clips, and outputs. These are framed as live or near-live executions showing the orchestrated swarm (specialised agents debating/fact-checking), empirical grounding (Google leak signals, Quality Rater simulation), compliance layers (Ethics Office/Meg), and human oversight.

Agency: 1st Live Demo

AEO (Answer Engine Optimisation) via SEO in ten minutes with one prompt.

Agency: 2nd Live Demo

Technical SEO audit + Quality Rater Review, based on official + leaked Google guidelines, completed in ten minutes (vs. a full day manually).

Agency: Demo 3

SEO analysis of a site grounded by 2023 HCU. Uses attributes/scores from the Google leak to simulate Google’s Quality Rating (processed in ~10 minutes vs. days of work). Anderson notes his 2025 leak/books were “training material” for this.

Agency: Demo 4

Further AEO via SEO demonstration in 10 minutes.

Additional posts: Roundup outputs (e.g., on HCU relevance in 2026), discussions of full agentic system (site audit → priority identification → content creation → human CEO sign-off/publish), and topical cohesion module plans. One post directly ties Agency to the academic paper: “When a Google paper describes exactly what you’ve built… almost a blueprint for what Shaun calls his ‘multi-agent digital marketing agency.’”

These X demos emphasise Agency as a cyborg exoskeleton (powerful strategy multiplier, not fully autonomous): Human provides vision/final approval; built-in Compliance Layer (Lawyer, Compliance Officer, Meg — Ethics Manager with zero-tolerance for bias/missing claims/E-E-A-T failures) guards against slop; heavy use of leak-derived verifiable signals + human empathy.

  • March 27, 2026: Updates to “Codename: Agency – By Your Command” and “Codename: Agency – A Weapon of Mass Disruption?” posts/articles, with X announcements. Anderson questions the disruptive potential of such tools in one person’s hands while stressing responsible use.

Ongoing (Late March 2026 onward)

Anderson’s trajectory—2023 skills vision → 2025 Agentica + deep Google leak forensics (contentEffort, E-E-A-T mapping) → Searchable cyborg apparatus → SMX Paris keynote → public X demos of Agency (speed, leak integration, governance in action)—provides a living, searchable practical counterpart to the academic paper’s plural/social/hybrid intelligence framework.

  • Core shared theme: The intelligence explosion is already happening in recursive agent ecologies and centaur workflows. AI swarms excel at rapid, specialized execution (as shown in the 10-minute demos), but human strategic vision, taste, empathy, E-E-A-T oversight, and institutional checks (like Meg’s Ethics Office) remain the critical leverage and safeguard.

This timeline reflects Anderson’s transparent, public output on hobo-web.co.uk and @Hobo_Web (where demos are shared with media like screenshots/videos for verifiability)

Disclaimer: Agency is a simulation. AI is a simulation! Some people think you are a simulation. Not many people know this aspect of AI. AI cannot do a lot of things people say it can do. For transparency, I need to say it is a simulation. For instance, I have an accountant baked into Agency. I am not an accountant, though, and neither is AI. This data should be reviewed BY YOUR REAL ACCOUNTANT, with the point being you have saved a lot of time collecting, sorting and reviewing the data before the real accountant reviews it. That is the essence of Agency. It is the essence of AI HITL Marketing. AI does the grunt work, the CEO signs the job off, and publishes under their credentials. This article was created using Grok, CLaude and Gemini to give a thoughtful, transparent overview of what Agency, and Agentic systems like it represent over the coming year. For me, at this point, the cyborg apparatus I predicted in my AI SEO ebook last year is now effectively built, and it is just a job of making it slicker and ensuring it is wielded correctly and responsibly by the user.

Hobo
Powered by  GDPR Cookie Compliance
Privacy Overview

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.