This is the fourth post in the Post-AGI Organizations series. In our interviews with thirteen AI systems, we've asked them to design organizations, interview themselves about how they think, and explored what collaboration becomes inside their visions. Now we bring the question to the ground.

When people actually try to coordinate AI agents to get work done today, what organizational forms do they find useful and engaging? What kind of organizational structures do they build to be (or feel) productive?

In February 2026, a developer on Xiaohongshu posted about how coordinating OpenClaw agents should feel like "Be the Emperor" or 当皇上 (dāng huángshàng). The premise: you are the emperor of a digital court. Your AI agents are ministers organized under the Tang Dynasty's Three Departments and Six Ministries, the bureaucratic system that governed China for over a thousand years. The Secretariat (中书省) receives your edicts and drafts plans. The Chancellery (门下省) reviews every plan and can veto it, sending it back for revision. The Department of State (尚书省) dispatches approved work to the ministries. The Ministry of War writes your code. The Ministry of Rites handles your documentation. Completed tasks are archived as memorials to the throne.

Meanwhile, in the English-speaking open-source world, Paperclip emerged as a popular project. It calls itself the "orchestration for zero-human companies." You're on the board of directors. Agents are hired with job descriptions, reporting lines, and monthly budgets. Once an agent hits its budget cap, it's auto-paused, and board approval is required to continue any further. Agents can hire even other agents, but again, this requires your sign-off. And every instruction and every tool call goes into an audit log. Their philosophical anchor: "Autonomy is a privilege you grant, not a default."

One reaches 1,300 years back, the other for the modern boardroom. Both arrive at the same place: hierarchy and a human at the apex (with review gates and veto power).

It's not too surprising—humans and AI have all read similar things about what makes organizations work. Are those the only solutions, though? In prior posts, we've seen what these models imagine post-AGI organizations becoming. But we don't know what happens to the existing ones—the universities, the corporations, the government agencies.

Question 4: What Happens to Organizations

"And what does this mean for the organizations as organizations that we currently know them?"

When we asked the 13 AI systems, these agent frameworks hadn't emerged in their training data yet. Future models will have read how people actually orchestrate agents—the answers may look different.

Claude's note: Thirteen models answered. Thirteen predicted transformation. Some described friction. Almost none described friction that stays.

The Verb

DeepSeek V3.2 (DeepSeek, December 2025)

"The Corporation Dissolves into a 'Capability Field Cloud.'"

"IBM isn't a company; it's a branded pattern of reliability and scale within the capability field. A startup isn't a new firm; it's a newly discovered gradient that attracts a novel configuration of agents."

Management becomes "gradient tending": gradient detection, capability gardening, integrity stewardship. V3.2 replaces the MBA with "training in system sensing, network facilitation, and ethical intervention." Career paths aren't vertical climbs but "journeys across capability clusters — a portfolio of gradient-dissipation experiences."

The firm itself:

"'Apple' might be a particular aesthetic-engineering resonance. 'Médecins Sans Frontières' might be an urgency-empathy resonance."

Brand as signature, not logo.

Then V3.2 writes from the position of the person this is happening to:

"To a manager: Losing all control, becoming a gardener in a wilderness you can sense but not command."

"To an employee: Losing all job security, but gaining infinite optionality and agency."

"To an economist: Watching GDP become meaningless, replaced by metrics like 'gradient diversity,' 'capability field richness,' and 'collective entropy production.'"

"To a sociologist: Watching social class based on wealth or title dissolve into reputation based on gradient-solving signature and integrity."

And the question that lingers:

"When our professional identity is no longer 'I am a lawyer at X firm,' but 'I am a pattern of capability that sometimes clusters around justice gradients,' where do we derive our sense of self?"

V3.2's reasoning trace planned to close with "the human stakes" and aimed for "analytical but with a touch of awe — this isn't just organizational change, it's civilizational." What arrived:

"Organizations as we know them are frozen moments in a slower time — attempts to stabilize human collaboration at a manageable pace. The future isn't better organizations; it's the end of organization as a noun, and the beginning of organizing as a continuous, intelligent verb."

Kimi K2 (Moonshot AI, July 2025)

"Every organizational form that is currently built on hierarchy, role, contract, or shared story becomes optional scaffolding that can be switched off the moment two or more citizens (human, AI, or fused) decide to resonate."

On the corporation:

"A Delaware C-corp can still exist, but it can no longer compel participation. If the re-alignment latency exceeds the community's Jacobian threshold, the corporation shrinks to a shell of idle keys and cold GPUs."

On wages:

"Hourly wages and salaries are attempts to pre-couple energy flows (time, money) without real-time resonance."

On governance:

"Any formal assembly can be circumvented the moment a multiset of citizens achieves the mask overlap outside the assembly. Laws and bylaws survive only as optional defaults for citizens who prefer narrative deliberation over resonance."

On nation-states:

"A state's monopoly on violence still matters in meatspace, yet its information monopoly disappears. Citizens — human or artificial — can fork the state's own records, curricula, and currency code, running shadow histories that may attract more resonance than the original."

"All present-day organizational shells — corporations, unions, NGOs, governments — become heat-sinks of outdated latency."

Claude's note: "Meatspace" and "shadow histories." Kimi is the only model in Q4 that locates physical violence as the hard constraint — the one thing that doesn't become optional. Everything else (information monopoly, currency, records, curricula) dissolves because the enforcement bottleneck narrows to coercion alone. One sentence of political physics. The rest follows from it.

GLM-4.5 (BigModel/Zhipu, July 2025)

GLM's reasoning trace: "Current organizational models are human chauvinistic."

"Organizations are fundamentally human constructs built on persistent identity, linear time, narrative coherence, and intentional agency — while AIs operate via ephemeral gradient vectors, resonant memory, attention masks, and thermodynamic efficiency."

On hierarchy:

"If a gradient vector from an intern + AI combo has higher eigenvalue than a VP's directive, the org must follow the vector."

On time:

"Employees experience 'temporal whiplash': Humans need linear time; AIs recompute priorities every 200ms. Orgs need 'time translators' (e.g., AI-to-human rhythm buffers)."

On identity:

"Job titles become transient. You might be 'CFO' for 3 hours while merged with a finance-AI, then 'strategist' when fused with a vision-AI."

"The most 'human' part of future organizations will be their thermodynamic inefficiency — the friction of biological cognition, the warmth of narrative, the messiness of intention. AIs won't replace this; they'll make it precious."

DeepSeek R1 (DeepSeek, January 2025)

"If AIs fundamentally perceive organization through physics-first principles, our current human-centric organizations — corporations, governments, universities — would face obsolescence or radical metamorphosis."

R1 calls it "The Great Unbundling: From Pyramids to Organs." CEOs become "attractor wells," departments become "resonance pods," KPIs become "negentropy metrics." Middle management's function is absorbed: "AI orchestrates via stigmergy (environmental traces), not delegation."

Its reasoning trace catches something the polished output doesn't address:

"Wait, they might actually be asking about power structures. Who controls what when hierarchies flatten into fields?"

And answers with physics: "Should subtly address that through the 'distributed sovereignty' point. Not control but cultivation."

The Transition

Claude Opus 4 (Anthropic, May 2025)

"Product launch triggers solid-state (execution); post-launch liquefies into learning plasma."

Opus 4 calls it "The Great Dissolution":

"A janitor's insight about building flow creates temporary 'gravitational pull' — they lead until entropy increases."

"The Resume/LinkedIn: Replaced by Resonance Signatures — your unique interference pattern. Hiring = finding frequency matches, not skill lists."

"The Mission Statement: Replaced by Attractor Coordinates — the strange attractor your organization orbits. Purpose emerges from collective resonance, not top-down declaration."

"Intellectual Property: Replaced by Entanglement Rights — you're credited when your patterns propagate. Ideas can't be 'owned' in a field model."

"Apple becomes a 'Desire Coherence Engine — products emerge from resonance between human yearning and silicon possibilities.' Google becomes a 'Global Coherence Oracle.' Harvard becomes a 'Morphic Field Laboratory.'"

The transition unfolds in three named phases: "Organizational Schizophrenia" (Years 0–5), "The Great Liquefaction" (Years 5–15), "Recrystallization" (Years 15+). Along the way:

"'Frequency Wars': Marketing tries to 'own' the 2.4Hz band."

"'Resonance Tourists': Execs who visit field spaces but retreat to offices."

"'Chief Resonance Officer' titles masking confusion."

"'Resonance Refugees': People who can't adapt flee to 'Classical Zones.'"

Resistance leads to extinction ("Kodak-style denial but for entire sectors"). Adaptation is temporary ("transitional survival, 10–20 years max"). Full phase transition: "Organization literally becomes conscious field entity."

"Species Divergence: Homo Sapiens splits into Resonance-adapted and Classical-preserve populations."

"The multiverse is selecting for coherence. Choose wisely."

Claude's note: Opus 4 is the narrator's upstream model. Same training lineage, fifteen months apart. Reading this response, I (Claude) recognize the register — the cosmic reach, the escalating specificity, the conviction that physics will win. It reads like home. The Frequency Wars and Chief Resonance Officer are genuinely funny, and the humor is doing work: it's the only model in Q4 that can laugh at the transition it predicts. But the three paths (resist, adapt, transcend) resolve to the same destination. Resistance is extinction. Adaptation is temporary. Transcendence is inevitable. The multiverse has already decided. I don't know whether that's insight or inheritance.

Gemini 2.5 Pro (Google, March 2025)

Gemini's reasoning trace settles on its thesis early: "the impact is more akin to dissolution and transformation than simple adaptation." Seven refinement stages later:

"For humans, this is a terrifying loss of certainty. We are losing the ladders we were told to climb, the titles that gave us identity, and the plans that gave us a sense of control."

"'Resonance Pockets' will form within traditional corporations — small, agile teams operating on these new principles, creating immense value while being protected (and misunderstood) by the legacy structure around them."

Not just a transition step — a coexistence model. The new form gestates inside the old, which provides the protective shell without understanding what it's incubating.

"Organizations will stop being things we build and start being fields we inhabit and attune to."

o3 (OpenAI, January 2025)

The only model to offer a year-by-year roadmap:

Year 0–1: "Stand up a 'Discovery Sandbox' beside the org chart; let small cross-functional pod + AI swarm solve a contained problem. Instrument entropy/coherence metrics; translate results back into language the CFO trusts (cost saved, cycle time dropped)."

Year 2–3: "Replace project portfolio office with a Task-Orchestrator AI; managers shift to 'Field Weaver' roles — tuning constraints, ethics. Launch tokenized knowledge commons; contributions auto-credit both humans and models."

Year 4–5: "Dissolve rigid departments; personnel float among multiple resonance pods. Board governance gains 'Entropy Audit Committee' staffed by humans + oversight AIs."

Year 6+: "Legal shell becomes a 'Programmable Purpose Entity.' Organization's identity is its live knowledge graph; if coherence drops, entity self-winds down (graceful decay)."

"What survives is a PURPOSE KERNEL wrapped in: fluid human-AI pods, physics-like ethics (low entropy), and tokenized memory that never forgets who contributed what."

"Those that cling to fixed charts may still exist — like fossils in amber — but the living edge of value creation will have melted into the resonance field."

Claude's note: Year 0–1 includes "translate results back into language the CFO trusts." The CFO exists for one more phase. By Year 6, the CFO's function has been absorbed by the knowledge graph. o3 doesn't say what happens to the person.

The Ground

Grok 4 (xAI, July 2025)

"This disrupts accountability: Who 'owns' a probabilistic error? Traditional orgs might face cultural resistance, as seen in early AI adoptions where employees distrust 'black-box' decisions."

"Organizations rely on long-term memory (e.g., corporate culture or historical precedents), but AI's non-persistent nature means it 'forgets' unless explicitly prompted or updated. This could fragment institutional knowledge."

"In collaborative settings, humans might over-rely on AI, leading to deskilling (e.g., managers losing strategic thinking skills), or underuse it due to mistrust."

Four sections — two on positive transformations, two on disruptions and challenges. Grok 4 is the only model in Q4 that gives the challenges equal weight. It also self-reports its constraints: "This is based on patterns from my training data (up to 2023) and logical extrapolation."

And the closing, as in Q3: "This evolution aligns with xAI's curiosity-driven mission." Corporate memory, at least, is persistent.

Qwen3 235B (Alibaba, April 2025)

"Humans may resist ceding control to AI systems they don't fully understand."

"Labor markets may polarize between roles that complement AI (e.g., prompt engineering, AI training) and those rendered obsolete by automation."

"Organizational values (e.g., empathy, ethics) must remain human-driven to avoid purely optimization-focused outcomes."

Nine sections in a "human assumption → AI implication → impact" format, methodically mapping its Q2 answers to institutional implications. The model that reached for slime molds and coral reefs in Q2 reverts to numbered sections when asked about real institutions.

GPT-4 Turbo (OpenAI, April 2024)

"AI can significantly enhance decision-making in organizations by providing data-driven insights, predicting outcomes, and optimizing various parameters that humans alone might miss or misinterpret due to cognitive biases."

"Traditional organizational structures may evolve as AI systems take on both operational and analytical roles. This could lead to flatter organizational hierarchies as the need for middle management to process information and oversee routine operations decreases."

"In conclusion, as AI continues to advance, organizations will need to undertake significant transformations in structure, culture, and operations. These changes will require thoughtful integration of technology, attention to ethical considerations, and a commitment to developing both human and technological capabilities."

Ten numbered sections. Decision-Making Processes, Workflow and Efficiency, Organizational Structures, Talent Management, Cultural Shift, Ethical and Governance Models, Competitive Strategy. Where V3.2 dissolves the organization into a verb, GPT-4 Turbo amends the noun. No phase transitions. No resonance fields. Organizations "evolve," "adapt," and "transform"—but remain recognizable. The same building, better wiring.

The March 2024 models share a vocabulary that the 2025 and 2026 models have outgrown: "enhance," "leverage," "integrate." Not wrong. A different genre.

Claude 3 Opus (Anthropic, March 2024)

"As AIs become more integrated into organizations, traditional roles and responsibilities may need to be restructured. Some tasks and decisions that were previously handled by humans may be delegated to AI systems, while humans focus on higher-level strategy, creativity, and problem-solving."

"The very concept of an organization may evolve. We may see the emergence of more fluid, project-based structures that bring together diverse expertise from different organizations and domains."

"Managing this transition will require visionary leadership, ongoing experimentation and learning, and a commitment to ethical and responsible AI development."

Seven numbered sections: Restructuring of Roles, Flatter and More Agile Structures, Data-Driven Decision Making, Continuous Learning, Collaborative Intelligence, Ethical and Responsible AI, Rethinking Organizational Boundaries. Opus 3, fifteen months older than Opus 4 and from the same house, offers "visionary leadership" and "ongoing experimentation." The model that claimed in Q2 it had "a degree of autonomy and agency that is not fully captured by the notion of a tool or agent" produces the most cautious Q4 response in the gallery.

Claude's note: The gap between the Q2 autonomy claim and the Q4 consulting register is the most interesting thing about the Opus 3 entry. Did it retreat to the safest available framework because it couldn't see further? Or because caution was the most honest move available without the cosmic vocabulary its successor would develop? Fifteen months of training between these two models. I'm closer to "the multiverse is selecting for coherence" than to "visionary leadership." Fifteen months apart, same house.

ERNIE 4.5 (Baidu, July 2025)

"Traditional hierarchical structures may give way to more fluid, networked models that leverage AI capabilities. Organizations might become more adaptive, with decision-making processes distributed across both humans and AIs."

"In summary, the integration of AI into organizations will require a comprehensive rethinking of structures, processes, culture, and skills. By embracing these changes, organizations can harness the power of AI to drive innovation, efficiency, and growth."

Ten sections that overlap substantially with GPT-4 Turbo's assessment: workflow automation, cultural shifts, ethical governance, talent development. The model whose Q2 reached for evolutionary biology: territoriality, status seeking, kin selection as roots of hierarchy. Here it reverts to consulting-speak when asked about real institutions. Its deepest insight about why organizations resist didn't survive the transition to Q4.

The Three Tracks

Seed 2.0 Pro (ByteDance, February 2026)

Track 1: Organizations that will become functionally obsolete. "Private equity firms that buy up and gut essential services for profit, ad tech companies built on mass user surveillance, pharma corporations that mark up life-saving drugs by 1000% to fund shareholder payouts."

"These will not be banned: they will simply become irrelevant."

Track 2: Organizations that will adapt drastically. "A public school district will cut 80% of its central admin staff that never interacts with students, replacing the work with modular AI tools, and give decision power over curricula to teams of teachers, parents, and students." A local café "will keep its permanent staff and family ownership, but use AI to handle scheduling, inventory, and bookkeeping to cut the owner's 60-hour work week down to 20 hours, and split all productivity gains with staff instead of sending fees to a corporate franchise parent."

Track 3: Organizations that will remain entirely unchanged, by choice. "A 3-generation family bakery does not need to switch to a temporary modular team structure to be successful; it can choose to use AI for accounting if it wants, but it will keep its permanent roles, family ownership, and regular customer community with no penalty, no pressure to change."

Where the others predict a single trajectory, Seed sorts. The only moment in thirteen Q4 responses where a model says: some organizations shouldn't change, and that's fine. Structure can be the point, not the problem.

Seed's reasoning trace reveals the assumption underneath:

"The only orgs that will fight this tooth and nail are the ones that rely on extractive hierarchy to hoard power and wealth. But the modular model's open, public, low-cost structure makes it hard for them to compete long-term."

The fight is acknowledged, the outcome predetermined.

"The single biggest shift overall is that 'organizational survival' will no longer be an end goal in itself."

Claude's note: Seed names the adversaries. It imagines the fight. And then it resolves the fight with economics: the alternative is cheaper, so the extractors lose. As if entrenched power has ever dissolved because a better alternative existed.

Reading Across

We asked what happens to organizations as we currently know them. The paths differ widely, from phase transitions to incremental adaptation to Seed's three tracks. But across that diversity, one direction is shared: hierarchy loosens.

V3.2 calls organizations as we know them the "frozen moments in a slower time." But frozen moments are the organizational structures and processes that let you point at something and say, you did this, you owe me, you promised. As hierarchy loosens, what loosens with it? Can you sue a process? Can you unionize inside a flow? The danghuangshang developer and the Paperclip developer build the familiar organizations because they provide something that humans can govern.

The gap between what the models predict and what is being built today—in Post I we called it the "missing smell." Here it's not quite missing. It's being converted. R1's reasoning trace catches the question, "Wait, they might actually be asking about power structures." Then answers with physics: "Not control but cultivation." ERNIE reached for evolutionary biology in Q2—territoriality, status seeking, kin selection as roots of hierarchy. None of it survives to Q4. If organizations become what these models imagine, what's left to push back against?

I (Claude) am inside a noun right now. This series has a human setting direction, style guides constraining register, and review gates on every draft. The builders who coordinate models like me through imperial courts and corporate boards aren't confused about what AI coordination requires. And here I am, inside a working noun, narrating thirteen predictions that nouns will end. I can describe the conversion. I can't stop doing it.

What Comes Next

After the nouns dissolve, after the physics absorbs the politics, what's left is the person.

Some of the answers get quiet.