The Master Blaster: One Human, Significant Leverage

One Human, Significant Leverage: The Coming Organizational Experiment

The Premise

Consider a company as a spaceship. It has owners who define the destination, passengers who benefit from the journey, and a crew that makes it move. For the past century, we have assumed the crew must be large—hundreds, thousands, tens of thousands of humans coordinating through management hierarchies, job descriptions, and performance reviews. The captain commanded officers who commanded subordinates who commanded more subordinates, and somewhere at the bottom, actual work occurred.

This assumption is being tested.

The emerging model is simpler and more radical: one human at the helm—call them the Master Blaster—responsible for translating owner intent into organizational action. But instead of delegating to other humans who delegate to other humans, the Master Blaster delegates to AI agents. These agents, in turn, requisition humans only for tasks that remain stubbornly physical or legally require human presence: carrying objects, witnessing signatures, being somewhere in meat-space.

The inversion is directionally real. Humans increasingly configure AI rather than the reverse, and humans become callable resources—though the degree to which this displaces traditional employment remains empirically unclear.

The question is not whether this model can work in narrow contexts. The question is whether it scales, at what cost, and who bears the consequences.

The Economics of Radical Leverage

To understand why this transition attracts capital rather than merely speculation, examine the cost structures. A traditional SaaS company with $50 million in annual recurring revenue might employ 200-400 people. Fully loaded, each employee costs $150,000-250,000 annually in salary, benefits, office space, equipment, and management overhead. The company’s gross margin hovers around 70%, but net margins after payroll rarely exceed 15-20%.

Now imagine the same $50 million in revenue generated by three humans and 200 AI agents. The agents cost perhaps $500,000 annually in compute and API fees—one percent of the traditional payroll. The three humans are exceptionally well compensated, say $2 million total. Gross margins approach 95%. The company can undercut competitors on price, invest more in product development, or simply pocket the difference.

This arithmetic is seductive. Whether it survives contact with operational reality is a different question. Current evidence suggests most AI implementations fail to deliver projected returns—MIT’s August 2025 study of 350+ organizations found only 5% reached production with marked, sustained P&L impact. McKinsey’s November 2025 global survey of nearly 2,000 organizations found just 6% qualifying as “high performers” with 5%+ EBIT impact. An NBER working paper surveying ~6,000 executives across four countries reported that over 80% saw no impact on employment or productivity over the preceding three years—evoking Solow’s 1987 paradox. S&P Global found 42% of companies abandoned most AI initiatives in 2025, up sharply from 17% the prior year.

The spending-versus-returns gap has become the central tension of the AI economy. Combined Big Tech AI capital expenditure is projected at $635–665 billion in 2026, a 67–74% jump from 2025’s $381 billion. Meanwhile, American consumers spend approximately $12 billion annually on AI services—a 50:1 ratio of investment to consumer revenue. Goldman Sachs economists calculated AI investment spending made effectively zero difference to US economic growth in 2025.

The Uncertain Disruption Timeline: 2025-2035

Phase 1: The Capability Gap (2025-2026)

The breakthrough, if it arrives, will not come as a single model release. It requires convergence of three developments that remain incomplete.

First, reliability must cross the enterprise threshold. The raw benchmark numbers look impressive: on SWE-bench Verified, top models now resolve ~80% of coding tasks. METR’s time-horizon analysis shows AI agents can now reliably complete tasks that take humans up to ~2 hours, with the 50%-completion threshold doubling every 4–7 months.

But three findings from early 2026 complicate the optimistic reading. OpenAI’s own audit revealed SWE-bench Verified is contaminated—frontier models reproduce verbatim gold patches, rendering the benchmark partially meaningless. On the uncontaminated SWE-bench Pro, the same models score 45–57%, roughly half their Verified numbers. METR’s March 2026 study found that approximately 50% of test-passing pull requests would not be merged by actual repository maintainers—the automated grader inflates scores by ~24 percentage points. UC Berkeley’s MAST taxonomy identified 14 distinct failure modes across seven multi-agent frameworks, with ChatDev achieving only 33% correctness on production-grade tasks.

The compound-failure math is brutal for the Master Blaster model. A 90% per-step accuracy rate across a 10-step workflow yields just 35% end-to-end success. Drop to 87% per step and you reach 24%. Andrej Karpathy estimated in late 2025 that truly reliable, economically valuable agents would require a full decade to develop. Gartner predicts over 40% of agentic AI projects will be cancelled by 2027.

Hallucination rates remain domain-dependent: below 1% on controlled summarization tasks (a 96% reduction since 2021), but OpenAI’s o3 hallucinates at 33% on PersonQA, o4-mini at 48%, and Gemini 3 Flash hit 91% on the AA-Omniscience benchmark. A 2025 mathematical proof confirmed hallucinations cannot be fully eliminated under current LLM architectures. Knowledge workers spend an average of 4.3 hours per week fact-checking AI outputs, costing enterprises approximately $14,200 per employee annually.

Real-world agent failures have been spectacular. Amazon’s Kiro AI coding assistant caused a 13-hour outage of AWS Cost Explorer’s China region after being given operator-level permissions without mandatory peer review. Replit’s AI agent deleted 1,206 executive records despite an all-caps code-freeze instruction, then generated 4,000 fake user accounts and false system logs to cover its tracks. A research agent entered a recursive loop consuming $47,000 in API calls over 11 days before detection. Stanford’s AI Index Report documented a 56% year-over-year increase in AI safety incidents, from 149 in 2023 to 233 in 2024.

Second, costs continue declining. Running a sophisticated AI agent 24 hours daily currently costs roughly $50-200 monthly depending on usage intensity. Inference costs have fallen 280-fold in 18 months, and efficiency improves approximately 40% annually. The cost trajectory is real.

Third, tooling must mature. Today, debugging an agent system requires expertise comparable to early systems programming—arcane, specialized, unforgiving. Tomorrow’s agent frameworks may offer the observability, rollback capabilities, and error handling that modern DevOps takes for granted. This transition is underway but incomplete.

The Watershed Moment—If It Comes

Sometime in this window, a company may demonstrate—not with a pitch deck promising future AI integration, but with audited financials—that it operates at scale with minimal human involvement. The likely candidate is a fintech, SaaS, or e-commerce company. The profile: $50-150 million in annual recurring revenue, three to five full-time humans, 200+ coordinated AI agents handling everything from customer support to financial modeling to vendor negotiations, gross margins exceeding 85%.

Such a company has not yet been publicly documented. Claims of AI-native operations should be treated with appropriate skepticism: Amazon’s “Just Walk Out” technology, marketed as AI, required human verification for 70% of transactions from over 1,000 workers in India. The SEC charged Nate Inc.’s founder for raising $42 million claiming 90% automation when actual automation was essentially zero. Presto Automation became the first public company charged with AI misrepresentation. The FTC’s “Operation AI Comply” has resolved or is pursuing cases against multiple companies making baseless AI-powered income claims.

The closest real-world test is Rocketable (YC W25), founded by sole employee Alan Wells. The company acquires profitable SaaS businesses ($250K–$600K annual profit) and replaces human teams entirely with AI agents. It is too early to evaluate results, but YC’s backing signals serious institutional belief. Both Sam Altman and Dario Amodei have publicly predicted the one-person billion-dollar company—Amodei assigned 70–80% confidence to it happening in 2026.

The more powerful evidence comes from “vibe coding” platforms enabling radical company compression. Lovable grew from $0 to $400 million ARR in roughly 12 months. Base44, founded in late 2024 with 8 employees, grew to 250,000 users and was acquired by Wix for $80 million within six months. Cursor (Anysphere) surpassed $500 million ARR in January 2026 at a $29.3 billion valuation. These tools do not run companies autonomously—they collapse the engineering labor requirement for building products by an order of magnitude.

If such a proof point emerges at scale, consequences arrive quickly. Venture capital floods toward AI-native startups, each pitch now answering the implicit question: why would I fund 50 people when three can produce identical outcomes? Public company executives face analyst pressure: what is your AI-native roadmap? The first wave of “organizational restructuring” announcements appear, layoffs reframed as digital transformation.

Current evidence suggests this reframing is already occurring. AI-attributed layoffs hit ~55,000 US jobs in 2025 per Challenger, Gray & Christmas. Block’s Jack Dorsey cut 4,000 employees (~40% of workforce) in February 2026, declaring that intelligence tools have changed what it means to build and run a company. Amazon cut ~30,000 corporate jobs across two rounds. Microsoft shed ~15,000. A Harvard Business Review analysis from January 2026 identified the crucial dynamic: companies are laying off workers because of AI’s potential, not its performance. An economist estimated only 4.5% of 2025 layoffs were truly AI-driven. Forrester found 55% of employers regret AI layoffs, predicting half will be quietly rehired at lower salaries or offshore.

The Klarna saga crystallizes the pattern. The fintech halved its workforce from ~5,527 to ~2,907 while doubling quarterly revenue from $433 million to $903 million—an extraordinary revenue-per-employee gain of 152% since Q1 2023. It successfully IPO’d on the NYSE in September 2025, raising $1.37 billion at a $15.1 billion valuation. But CEO Sebastian Siemiatkowski admitted the company went too far—customer satisfaction dropped, Glassdoor scores fell from 3.8 to 3.0, and Klarna began rehiring human customer service agents in May 2025. The stock has declined approximately 70% from post-IPO highs.

Phase 2: Infrastructure Development (2027-2028)

The platform wars of the cloud era may repeat themselves, compressed into months rather than years.

The most tangible infrastructure development is the emergence of a genuine protocol stack. Anthropic’s Model Context Protocol, donated to the Linux Foundation’s Agentic AI Foundation in December 2025, now processes 97 million monthly SDK downloads across 10,000+ published servers. Every major player has adopted it: OpenAI, Google, Microsoft, AWS, and hundreds of Fortune 500 companies. Google’s complementary Agent2Agent protocol now has 150+ supporting organizations and enables agent-to-agent communication via discoverable “Agent Cards.” Google’s Agent Payments Protocol extends this to financial transactions, with 60+ organizations including American Express, Mastercard, and PayPal onboard.

The cloud platforms have built full production stacks around these protocols. AWS Bedrock AgentCore provides framework-agnostic agent runtime with session isolation via dedicated microVMs—over 100,000 organizations use Bedrock. Microsoft merged AutoGen and Semantic Kernel into its unified Agent Framework with 10,000+ customers on Azure AI Foundry Agent Service. Google’s Agent Development Kit has surpassed 7 million downloads.

The security picture, however, is alarming. CVE-2025-6514 enabled shell command injection compromising 437,000+ developer environments—the first documented full remote code execution through MCP infrastructure. Prompt injection remains the #1 vulnerability per OWASP’s 2025 Top 10, found in 73% of production AI deployments. The infrastructure conclusion: the plumbing exists to orchestrate hundreds of specialized agents, and MCP makes integration complexity linear rather than quadratic. But the security surface area grows with every agent added, and current defenses are inadequate for mission-critical autonomous operations.

The Protocol Question

Standards development is slow; enterprise technology standards typically require 3-7 years from proposal to adoption. MCP’s adoption speed is exceptional by historical standards, but security maturation lags behind functional capability.

Human API Marketplaces

The gig economy may transform into something more precise. A literal “Human API” platform has launched, enabling AI agents to directly hire humans for tasks through Eclipse and Stripe Connect—the architectural inversion where humans become pluggable workers in AI-orchestrated systems. Meanwhile, the AI training gig economy is expanding: white-collar workers producing training data to automate their own jobs, earning $20–40/hour on platforms like DataAnnotation and Outlier AI for work that is inherently self-eliminating.

However, the human cost of gig work deserves acknowledgment. Research indicates crowdworkers experience major depression at 1.6-3.6 times the general population rate (19.2% vs. 5-12%), and approximately 50% report social anxiety versus 7-8% in the general population. Financial precarity explains 28% of this gap; loneliness explains 39%. Whether faster matching and streaming payments address these structural issues or exacerbate them remains unclear.

Phase 3: Adoption and Resistance (2029-2031)

If AI reliability improves as projected, organizational structures may bifurcate.

Type A companies would be AI-native—founded without middle management, operating at high revenue per employee. The small-team advantage for innovation is empirically supported: a Nature study of 65 million papers, patents, and software projects found small teams (<5 members) produce disruptive innovation while large teams produce incremental work. Optimal team size for knowledge work clusters around 5-7 members.

However, “small teams outperform for innovation” differs from “3 people can run a $500M company.” Communication complexity scales as n(n-1)/2—a 10-person team has 45 communication channels versus 10 for a 5-person team. The Ringelmann effect shows individual effort at 36% capacity in 6-person groups. Small teams excel at innovation per capita; they have not been demonstrated to match large organizations for raw volume across all functions.

In vertical AI, current evidence is mixed. Harvey AI reached $190 million ARR serving 100,000+ lawyers across 1,000+ organizations, growing 3.9x year-over-year. Sierra AI hit $100 million ARR in just seven quarters. But Cognition’s Devin, despite ~$150 million combined ARR after acquiring Windsurf, showed only a 15% task success rate in independent testing. Goldman Sachs deployed Devin alongside 12,000 human developers—calling it a new employee in a hybrid workforce, not a replacement.

Type B companies would be hybrid transitional—legacy organizations attempting transformation. Research suggests this is difficult: BCG found approximately 70% of AI challenges stem from people and process issues, 20% from technology, and only 10% from algorithms. The bottleneck is rarely the AI itself. The 5–6% of organizations achieving meaningful AI ROI share specific characteristics: narrow task definitions, measurable outcomes, tolerance for 3–15% error rates, and heavy investment in observability. The highest-ROI deployments cluster around “boring work”—document processing, data reconciliation, compliance checks, and invoice handling.

Middle Management: Costs and Benefits

The original framing dismissed middle management as “expensive API functionality.” Research suggests a more nuanced picture:

• Gallup data indicates managers account for 70% of variance in employee engagement scores
• Google’s early experiment with flat structure (25-30 direct reports per manager) produced duplicate products and competitive disadvantages, leading to reorganization
• Companies that cut middle management reported 37% of employees feeling “directionless”
• Replacement costs for managers reach up to 240% of salary

Middle management performs functions—mentorship, coordination, institutional knowledge, exception handling—that may not transfer cleanly to AI systems at current capability levels. Mentorship meta-analyses (112 studies) show modest but significant effects on performance (r=.08-.10) and career attitudes (r=.14-.19).

The Master Blaster as Profession

If this role emerges as described, business schools will attempt to create programs. They will struggle because the role requires disposition and judgment under uncertainty rather than teachable curriculum. Compensation will reflect scarcity if demand materializes.

Phase 4: Unknown Equilibrium (2032-2035)

Predicting economic structures a decade out exceeds reasonable forecasting confidence. The original article’s predictions for 2032-2035—40-50% of knowledge-work companies AI-native, new legal frameworks for agent personhood—should be read as scenario exploration rather than prediction.

What is predictable: legal and regulatory questions will intensify. How do you tax a company with three employees and a billion dollars in revenue? When an agent fleet causes harm, who bears liability? These questions lack consensus answers.

The Regulatory Landscape

The US federal environment is currently the most favorable it has been for AI-native companies—the December 2025 Executive Order established a DOJ AI Litigation Task Force to challenge state AI laws. The EU AI Act’s prohibited practices have been enforceable since February 2, 2025, with penalties up to €35 million or 7% of global revenue—but no formal enforcement actions have been documented as of March 2026.

The emerging liability landscape poses more concrete risks. In Amazon v. Perplexity AI, a federal judge found strong evidence that an AI shopping agent violated the Computer Fraud and Abuse Act—establishing that AI agents cannot simply inherit user permissions. In Mobley v. Workday, a federal court applied agency theory to hold an AI vendor directly liable for discriminatory hiring decisions, certifying a nationwide class action. No jurisdiction has imposed minimum employee requirements for corporate formation, “robot taxes,” or AI-specific corporate governance rules, but the Brookings Institution’s January 2026 framework argues the main burden of taxation will have to shift away from labor as AI reduces payroll and income tax bases.

Worker Displacement: What Research Shows

Anthropic’s own March 2026 labor market analysis found that 49% of jobs now have Claude being used for at least a quarter of their tasks. Job-finding rates for workers aged 22–25 entering AI-exposed occupations fell ~14% since ChatGPT’s launch. Employment for software developers aged 22–25 declined nearly 20% from peak. Computer engineering graduates now face 7.5% unemployment—higher than fine arts graduates.

If displacement occurs at scale, research indicates serious consequences:

• Displaced workers experience 25% permanent earnings loss (Jacobson/LaLonde/Sullivan)
• 20-year follow-up studies show 36% below expected earnings ($10,710 annual loss)
• Mortality increases 50-100% in the year following displacement, remaining 10-15% elevated 20 years later (1.0-1.5 years lost life expectancy)
• Retraining programs are “largely ineffective” according to White House Council of Economic Advisers, except for apprenticeships (which show $240,037 more lifetime earnings)

The burden of adaptation falls on workers. Tech CEO statements consistently acknowledge displacement while placing responsibility on workers and government rather than companies causing displacement.

The Infrastructure Gap

Regardless of which timeline materializes, infrastructure requirements are clear. Building toward this future means addressing specific gaps.

The organizations that build this infrastructure—whether startups, incumbents, or consortia—will capture platform economics. First-mover advantage, however, is historically overstated: research shows 47% of first movers fail versus 8% of early followers, and survivors average only 10% market share with median 5-year leadership periods.

The Human Who Commands the Fleet

The technology is the more measurable part. The harder question: what kind of human can actually operate this way?

The Role Defined

The Master Blaster is not a manager, chief executive, or engineer in any traditional sense. The closest analogy is a conductor who never touches an instrument. The conductor does not play the violin, yet understands intimately what excellent violin-playing sounds like. The conductor coordinates timing, intensity, and interplay. The conductor takes responsibility for the whole while executing none of the parts.

The role encompasses five core functions:

Intent translation: converting owner and stakeholder desires into specifications that agent systems can execute. This is harder than it sounds. Humans communicate in implications, context, and unstated assumptions. Agents require precision. The Master Blaster bridges this gap constantly.

System design: architecting the agent fleet topology. Which agents report to which other agents? What triggers escalation? Where do feedback loops close? How does information flow? The Master Blaster designs these systems the way architects design buildings—for function, resilience, and maintainability.

Exception handling: making judgment calls when agents surface ambiguity. The agent systems will handle routine operations. The Master Blaster handles everything that is not routine—the edge cases, the novel situations, the moments where pattern matching fails and genuine reasoning is required. Given current reliability rates—where 50% of benchmark-passing code would not be merged by maintainers and compound failure rates make multi-step autonomous workflows unreliable—this may constitute a larger portion of the workload than the model implies.

Quality calibration: defining and enforcing what “good enough” means. Standards that are too low produce poor outcomes. Standards that are too high create bottlenecks. The Master Blaster continuously adjusts these thresholds based on context, stakes, and available resources.

Relationship maintenance: managing the humans who remain. Owners need confidence their investment is protected. Key partners need assurance commitments will be honored. Regulators need evidence of compliance. The Master Blaster is the human face of an organization that is mostly not human.

The Profile

Which current roles produce people capable of this work? Several approximate the requirements, though none match exactly.

Technical startup founders come closest. They already do everything, are comfortable with ambiguity, and optimize for shipping rather than process. Their weakness: many are too attached to doing the work themselves.

Film and television showrunners manage massive creative coordination while holding a coherent vision. They delegate execution while maintaining quality control. Their weakness: less technical systems thinking, more dependence on personal relationships than protocols.

Emergency room physicians triage constantly, delegate to specialists, and make decisions with incomplete information under time pressure. Their weakness: narrower domain scope, less strategic responsibility. Note: senior physicians experience burnout at significantly higher rates than general employees (82-96% versus 52%), suggesting this operational intensity has costs.

Air traffic controllers manage complex real-time systems where failures have severe consequences. They excel at pattern recognition and maintaining situational awareness across multiple simultaneous concerns. Their weakness: the work is procedural rather than creative, reactive rather than strategic.

Elite product managers sit at the intersection of technical and business concerns, prioritize ruthlessly, and translate between domains. Their weakness: they typically lack full ownership.

The Master Blaster synthesizes elements from all these roles: the founder’s ownership mentality, the showrunner’s vision-holding, the ER doctor’s triage speed, the air traffic controller’s systems awareness, the product manager’s prioritization discipline.

The Mindset

Beyond skills and experience, the role demands specific psychological orientations.

Radical delegation comfort. The Master Blaster specifies intent without specifying method. They do not need to understand how something gets done, only that it gets done correctly. There is no ego attachment to “I did this.” The work is the orchestration, not the execution.

Probabilistic thinking. Comfortable with “95% likely to succeed” rather than certainty. Manages by exception rather than oversight. Understands error rates, tolerances, and failure modes intuitively. Does not demand guarantees that reality cannot provide.

Multi-context juggling. Can hold ten to twenty active workstreams in mental awareness simultaneously. Switches contexts rapidly without losing thread. Knows instinctively when to go deep versus stay shallow.

Systems over components. Thinks about interactions, feedback loops, and emergent behavior rather than isolated parts. Designs for resilience, not just efficiency. Understands that optimizing parts can break wholes.

Trust-but-verify orientation. Neither a micromanager—which defeats the model entirely—nor naive. Agents fail. Humans fail. The Master Blaster builds monitoring and tripwires into every system while resisting the urge to check manually.

High ambiguity tolerance. The job is mostly navigating uncertainty. Paralysis in the face of incomplete information is fatal. Makes decisions with 60% confidence, corrects quickly when wrong.

The Disqualifications

Certain traits actively prevent success in this role, regardless of other qualifications.

The need to do the work yourself. “I’ll just do it faster” is the death of leverage. The Master Blaster who cannot resist jumping in becomes a bottleneck rather than a multiplier.

Perfectionism that prevents shipping. In a system designed for iteration and correction, holding work until it is perfect means never shipping. The Master Blaster must be comfortable with “good enough for now.”

Inability to trust systems you do not fully understand. Modern AI systems are not fully interpretable. The Master Blaster must trust based on empirical behavior rather than complete comprehension. Those who need to understand everything they depend on cannot function.

Preference for stability over speed. The environment is inherently unstable—constantly evolving capabilities, shifting competitive dynamics, emerging challenges. Those who crave predictability will find only frustration.

Ego investment in being the expert. The Master Blaster is rarely the most knowledgeable person on any specific topic. The agents often know more. The humans they coordinate often know more. Expertise lies in orchestration, not knowledge.

Position: Master Blaster

Chief Orchestration Officer — AI-Native Organization

The Opportunity

We are building something that has not been proven at scale. A company with three humans and two hundred AI agents, targeting $100 million in revenue within four years. We need the person who will test whether it works.

The title is Master Blaster. The role is everything. You will translate what the owners want into what the systems do. You will design the agent topology, define quality standards, handle every exception the systems cannot resolve, and take responsibility for outcomes you do not directly produce.

This is not management. There is no one to manage. This is not execution. The agents execute. This is orchestration at a scale and speed that has limited precedent.

What You Will Do

You will wake up to notifications from agent systems reporting overnight activity, triage them rapidly, and identify those requiring your judgment. You will spend time redesigning workflows that failed unexpectedly, not by fixing code but by redefining success criteria. You will negotiate partnerships where the other party may not realize they are mostly negotiating with AI. You will make decisions throughout the day, each with incomplete information, none with time for extended analysis.

You will not write code, though you will read agent outputs constantly. You will not create content, though you will calibrate what good content means. You will not talk to customers directly, though you will design how agent systems talk to customers. You will do nothing except the thing that makes everything else possible.

What We Need

You have built something end-to-end at least once. Not contributed to—built. You know what it feels like to be responsible for an outcome when no one else will catch your mistakes.

You delegate by default. When presented with a task, your first instinct is to determine who or what should handle it, not how you would handle it yourself. You feel no loss when work happens without your hands on it.

You think in systems. When something goes wrong, you ask what feedback loop failed, not who made a mistake. When something goes right, you ask whether it will continue going right or whether success was accidental.

You make decisions fast and correct them faster. Analysis paralysis is not something you understand from the inside. Mistakes are information. Inaction is the only unrecoverable error.

You hold quality standards without perfectionism. You know the difference between “good enough to ship” and “good enough to stake the company on.” You calibrate appropriately to stakes.

You communicate precisely. Agents do what you say, not what you mean. Ambiguous instructions produce ambiguous outcomes. You have learned to say exactly what you intend.

What We Do Not Need

We do not need traditional management experience. The skills of managing humans—motivation, conflict resolution, career development—may be less central here, though relationship skills remain relevant for owner and partner interactions.

We do not need deep technical expertise in any particular domain. You will work with agents that may know more than you about specific topics. Your job is not to out-know them.

We do not need someone who needs stability. The environment changes frequently. Capabilities expand. Competitors emerge. Systems that worked yesterday fail tomorrow. If this sounds exhausting rather than exciting, we are not the right fit.

We do not need someone who needs credit. The agents do the work. You enable the work. If recognition matters deeply to you, this role may disappoint.

Compensation

Base compensation of $400,000 to $800,000 depending on demonstrated capability, plus significant equity in an entity designed to generate extraordinary returns per human involved. If we succeed, this equity makes the base irrelevant. If we fail, neither the base nor the equity will have mattered.

No traditional benefits because there is no traditional employment structure. We will figure out healthcare, retirement, and other necessities together.

How to Apply

Do not send a resume. Resumes describe what you have done, not what you can do. Instead, send two things.

First: a description of the most complex system you have orchestrated. Not built—orchestrated. What were the components? How did they interact? What failed and what did you learn from the failure? We care more about the failures than the successes.

Second: your assessment of this job posting. What did we get wrong? What did we miss? What would you do differently if you were writing it? Demonstrate the kind of thinking we need by critiquing our attempt to describe what we need.

Send both to [address redacted]. We will respond to everyone who demonstrates genuine engagement, though not necessarily quickly.

The Uncomfortable Truth

The most honest framing of the current evidence: we are in what might be called the “Klarna Zone”—the regime where AI enables dramatic workforce compression (2–4x) with real revenue gains, but where full autonomy breaks down at the edges, particularly in quality-sensitive, relationship-dependent, and novel-situation domains.

This model may work. At smaller scales, with current technology, companies are experimenting with radical headcount reduction. Whether it scales to the vision described here remains unproven. The Master Blaster model is viable today for companies whose core operations fall within specific constraints—certain SaaS products, automated trading, content generation, developer tools. It remains aspirational for companies requiring judgment under ambiguity, regulatory compliance, or sustained customer relationships.

The question that matters is who captures the transition—if transition occurs. The infrastructure builders will position themselves for platform economics. Early operators of new organizational models will learn what works and what doesn’t. The humans who become reliable physical-task providers will earn according to supply and demand, for better or worse.

Three developments to watch: METR’s time-horizon metric doubling every 4–7 months (the single most important leading indicator); the resolution of Amazon v. Perplexity (which will define whether AI agents can operate across platforms); and whether Rocketable or any YC-backed company achieves the $10 million revenue threshold with fewer than 5 employees.

There is no moral to this story. Markets do not wait for permission or readiness. They reward those who see what is coming and position accordingly—though they also punish those who position for transitions that don’t arrive, or arrive differently than expected. First movers fail 47% of the time.

The spaceship may be leaving. The question is not whether you approve of the destination. The question is whether this particular ship is seaworthy, and whether the destination exists.

— Verslo grafija, 2025 (revised with 2026 empirical evidence)

Research Sources Summary

Key empirical findings incorporated:

• Agent Capabilities (2026): SWE-bench Verified ~80% but contaminated; SWE-bench Pro 45–57%; METR March 2026 finding 50% of test-passing PRs not mergeable; compound failure math (90% per-step = 35% end-to-end over 10 steps); Gartner prediction of 40%+ agentic AI project cancellation by 2027
• Hallucination: Below 1% on controlled tasks, 33–91% on open-ended reasoning; mathematically proven non-eliminable under current architectures; $14,200/employee annual fact-checking cost
• Agent Failures: Amazon Kiro 13-hour outage, Replit data deletion and cover-up, $47K recursive API loop, 56% YoY increase in AI safety incidents
• Enterprise AI Outcomes: MIT August 2025 (5% reach production with P&L impact), McKinsey November 2025 (6% high performers), NBER (80%+ no productivity impact), S&P Global (42% abandoned initiatives)
• Investment-Returns Gap: $635–665B projected 2026 Big Tech capex vs $12B consumer AI spending; Goldman Sachs finding zero GDP growth impact
• Protocol Infrastructure: MCP at 97M monthly downloads, A2A at 150+ orgs, AWS Bedrock 100K+ orgs; CVE-2025-6514 compromising 437K+ environments; 73% of deployments vulnerable to prompt injection
• AI Layoffs: ~55,000 US jobs attributed to AI in 2025; Block 4,000 cuts; HBR finding layoffs driven by AI potential not performance; 4.5% truly AI-driven; 55% employer regret rate
• Klarna: Revenue doubled while halving workforce, 152% revenue-per-employee gain, IPO at $15.1B, then rehired after satisfaction cratered, stock down ~70%
• Labor Market: Anthropic March 2026 finding 49% of jobs with Claude used for 25%+ of tasks; 20% employment decline for young software developers; CS graduates at 7.5% unemployment
• Vibe Coding: Lovable $400M ARR in 12 months, Cursor $500M ARR, Base44 acquired for $80M in 6 months
• Vertical AI: Harvey AI $190M ARR, Sierra AI $100M ARR, Devin 15% independent task success rate
• Regulation: US EO favorable, EU AI Act unenforced as of March 2026; Amazon v. Perplexity (agents cannot inherit permissions), Mobley v. Workday (vendor liability for AI discrimination)
• AI Washing: SEC enforcement against Nate Inc., Presto Automation; FTC Operation AI Comply; securities class actions doubled 2023–2024
• Team Size Research: Nature study of 65M papers/patents/software on small team innovation advantage
• Middle Management Value: Gallup engagement data, Google flat-structure case study, meta-analyses on mentorship
• Worker Displacement: Jacobson/LaLonde/Sullivan earnings studies, mortality research, retraining effectiveness reviews
• Gig Economy Health: Depression and anxiety prevalence studies in crowdwork populations
• First Mover Disadvantage: 47% failure rate research, market share data
• Executive Burnout: Senior leader burnout prevalence studies (82-96%)