research

Automation for Good: Why the Future Belongs to Those Who Build, Not Those Who Fear

"The best way to predict the future is to invent it." — Alan Kay

"Automation is not about replacing humans. It's about replacing tasks that humans shouldn't have to do." — Adapted from Clay Christensen

Abstract

The conversation about artificial intelligence and automation has become a monoculture of anxiety. Headlines oscillate between utopian promises of post-scarcity abundance and dystopian warnings of mass unemployment, algorithmic control, and existential risk. Both narratives share a common assumption: automation is something that happens to people. The builder — the person who writes the code, designs the system, deploys the tool — disappears from the frame. This think piece argues that the dominant fear-based framing of automation is not only incomplete but actively harmful, because it obscures the most important variable in the equation: who builds, and for whom. Drawing on real examples from open-source AI systems — Hermes Agent (autonomous content and research automation), piranhabomb (swarm-based code intelligence), and AIGuaratuba (local community intelligence) — we propose that automation is not a monolithic force but a tool shaped by the values of its builders. The question is not "will automation take your job?" The question is: "what kind of future are we automating — and for whom?" The answer depends not on technology but on who has the power to build.

The Fear Monoculture

We are living through the most automation-anxious period in modern history. The World Economic Forum's 2025 Future of Jobs Report estimates that 83 million jobs will be displaced by 2027, while 69 million new roles will emerge — a net loss of 14 million positions. Goldman Sachs puts the figure at 300 million jobs exposed to AI automation globally. McKinsey projects that by 2030, up to 30% of hours worked in the US economy could be automated. These numbers are not wrong. They are incomplete.

What's missing is the agency dimension. Every one of these projections treats automation as a weather system — something that arrives, displaces, and leaves. But automation is not weather. It is architecture. Someone designs it. Someone builds it. Someone deploys it. Someone profits from it. And someone bears the cost.

The fear narrative serves specific interests. When automation is framed as an unstoppable force of nature, the people who build it escape accountability. "AI will take your job" implicitly positions the technology as the actor and the worker as the passive recipient. But behind every automation system is a set of choices: what to automate, for whose benefit, with what safeguards, and under whose control. These are political choices disguised as technical inevitabilities.

This is not a new pattern. The Luddites of the 1810s were not anti-technology — they were anti-monopoly. They objected not to the stocking frame but to the fact that the stocking frame was controlled by factory owners who used it to suppress wages and eliminate craft autonomy. The technology was the instrument; the power structure was the problem. Two centuries later, the same dynamic plays out with AI: the technology is neutral; the distribution of power is not.

Part I: The Builder's Counter-Narrative

What Builders Actually Build

While the fear narrative dominates public discourse, a parallel world of building is producing a fundamentally different story about automation. These are not corporate AI labs optimizing for shareholder value. They are independent developers, open-source communities, and small teams creating tools that automate for people rather than instead of people.

The distinction matters. There is a categorical difference between:

1. Automation that extracts value — replacing human labor to reduce costs, concentrating gains in the hands of capital owners 2. Automation that amplifies intent — reducing the friction between a person's creative will and its execution, distributing capability to the builder and their community

The first model is what most people imagine when they hear "AI automation." It is the factory model: capital-intensive, centrally controlled, designed to minimize human participation in the value chain. The second model is what open-source builders are creating: tools that make individuals and small teams more capable, not less necessary.

Hermes Agent: Automating the Boring Parts of Thinking

Hermes Agent, developed by Nous Research, is an open-source autonomous agent framework. At first glance, it looks like another AI chatbot. But its architecture embodies a specific philosophy about what automation should do.

Hermes automates cognitive overhead — the research, synthesis, formatting, and coordination tasks that consume time but don't represent a person's core creative contribution. A journalist using Hermes can research a topic across dozens of sources in minutes rather than hours. A developer can generate boilerplate code, test suites, and documentation while focusing on architecture decisions. A researcher can synthesize findings from 50 papers into a coherent analysis without spending weeks on the mechanical work of reading and summarizing.

The critical design choice is that Hermes does not make decisions for the user. It does not choose what to write, what to publish, or what to believe. It automates the process while leaving the judgment to the human. This is the difference between a tool and a replacement. A word processor does not write your novel. A calculator does not solve your engineering problem. Hermes does not think for you — it thinks with you, handling the parts of cognition that are mechanistic so you can focus on the parts that are creative.

The practical result: a solo content creator running the blog.lermf.org pipeline — trend detection, research synthesis, article generation, image creation, multi-language translation, SEO optimization, and deployment — produces output that would require a team of 5-8 people in a traditional media organization. The automation did not eliminate the team. It made the individual equivalent to a team. That is amplification, not displacement.

piranhabomb: Swarm Intelligence for Code Quality

piranhabomb takes a different approach to automation: instead of amplifying an individual, it distributes intelligence across a swarm. Built on the concept of nullclaws — lightweight, sandboxed agent units that execute in WebAssembly isolation — piranhabomb runs millions of agent steps per second across thousands of parallel workers.

The practical application is code review and quality assurance. Traditional code review is a bottleneck: a senior developer reads pull requests, identifies bugs, suggests improvements, and approves or rejects changes. This process is slow, subjective, and limited by the reviewer's attention. piranhabomb automates the mechanical review — pattern matching, style consistency, vulnerability detection, dependency analysis — while escalating judgment calls to the human developer.

The swarm model is significant because it demonstrates that automation does not have to be centralized. piranhabomb does not have a single "brain" making all decisions. It has thousands of tiny agents, each contributing a small piece of intelligence, with the aggregate behavior emerging from their interaction. This is closer to how biological systems work — an ant colony does not have a central planner; it has thousands of simple agents whose interactions produce complex, adaptive behavior.

For the developer using piranhabomb, the effect is twofold: (1) code quality improves because automated review catches what human reviewers miss, and (2) the developer's time is freed for the creative work of designing systems rather than policing syntax. The automation handles the tedious; the human handles the meaningful.

AIGuaratuba: Intelligence That Belongs to the Community

AIGuaratuba represents perhaps the most radical reimagining of what automation can be. Rather than building a tool for individuals or corporations, AIGuaratuba builds local intelligence — AI systems that serve specific communities, learn from local knowledge, and operate under local control.

The name is deliberate: "Guaratuba" is a municipality in Paraná, Brazil, and the project embodies the principle that intelligence should be situated. A generic AI trained on the totality of the internet knows a little about everything and a lot about nothing that matters to any particular community. AIGuaratuba inverts this: it knows a lot about one place, one set of needs, one set of cultural contexts.

For a rural community in southern Brazil, this might mean an AI system that understands local agricultural conditions, municipal regulations, community health resources, and cultural practices — and can provide relevant, accurate, culturally appropriate information to residents who would otherwise have to navigate bureaucratic systems designed for urban populations. The automation does not replace community members. It gives them access to intelligence that was previously available only to those with the education, connections, or resources to navigate complex systems.

This is automation as democratization. Not "AI will replace your doctor" but "AI will help you understand what your doctor told you." Not "AI will eliminate local governance" but "AI will help you participate in it."

Part II: The Counter-Arguments — Taking Fear Seriously

A serious case for automation cannot dismiss the fears that accompany it. The fear narrative, while incomplete, is not wrong about everything. Three objections deserve careful engagement.

Objection 1: Job Displacement Is Real

The most concrete fear is economic: automation eliminates jobs, and the new jobs it creates may not be accessible to the people who lost the old ones. This is not theoretical. The US manufacturing sector lost 5.6 million jobs between 2000 and 2010, driven partly by automation. The Bureau of Labor Statistics projects that 1.4 million workers will be displaced by AI by 2030, with only 600,000 able to transition to new roles without significant retraining.

The builder's counter-narrative does not deny this. What it challenges is the inevitability framing. Jobs are not eliminated by technology; they are eliminated by policy choices about how technology is deployed and who benefits. Germany's Kurzarbeit program, which subsidizes reduced work hours rather than layoffs during economic transitions, has maintained employment levels through multiple automation waves. Denmark's flexicurity model combines flexible labor markets with robust retraining programs and social safety nets. These are not anti-technology policies — they are pro-human policies that shape how technology is deployed.

The open-source automation exemplified by Hermes, piranhabomb, and AIGuaratuba represents a third path: automation that creates new capabilities rather than eliminating existing positions. When a solo creator produces team-level output, the question is not "did a team lose jobs?" but "can more individuals now compete with teams?" The answer — so far — is yes. The automation lowered the barrier to entry rather than raising it.

Objection 2: Power Concentration

The second objection is structural: AI automation concentrates power in the hands of the few companies that control the models, the data, and the deployment infrastructure. OpenAI, Google, Anthropic, and Meta collectively control the frontier models that power most AI applications. Their decisions about what these models can and cannot do shape the boundaries of automated possibility for everyone.

This objection is largely correct for closed-source AI. The cathedral model — where a single entity designs, builds, and controls the system — naturally concentrates power. But it is not the only model. The open-source movement in AI represents a genuine alternative. Llama, Mistral, Qwen, and other open-weight models have democratized access to capable AI systems. Hermes Agent builds on this foundation: it is open-source, locally deployable, and customizable by its users.

The distinction between access and control is critical. A corporation that licenses GPT-4 has access but not control — OpenAI decides what the model can do. A developer running Hermes with Llama locally has both access and control — they can modify the model's behavior, add capabilities, remove restrictions, and deploy it in ways the original developers never anticipated. This is the difference between renting intelligence and owning it.

AIGuaratuba takes this further: it does not merely give individuals control over a generic AI system. It gives communities control over their own intelligence infrastructure. The power concentration objection is valid for the cathedral model of AI. It is structurally invalid for the bazaar model — provided the bazaar model is actually built and supported.

Objection 3: Alignment and Safety

The third objection is existential: AI systems that are sufficiently capable and autonomous may act in ways that are misaligned with human values, intentions, or survival. This is the "paperclip maximizer" scenario — an AI that pursues a seemingly benign goal with catastrophic intensity because it lacks the common sense to know when to stop.

This objection is legitimate and must be taken seriously. But it applies primarily to autonomous systems operating without human oversight. The automation paradigm described in this piece — Hermes, piranhabomb, AIGuaratuba — is explicitly human-in-the-loop. These tools automate tasks, not decisions. They handle the mechanical while leaving the judgment to people. The alignment problem is real, but it is most acute for systems designed to operate independently, not for systems designed to amplify human agency.

The deeper point is that alignment is not only a technical problem. It is also a political one. Who defines "aligned"? For whose values? With what accountability? The open-source approach does not eliminate alignment risk, but it makes alignment negotiable — each user, community, or organization can define and enforce its own alignment criteria rather than accepting the alignment choices of a corporate monopoly.

Part III: The Economics of Building vs. Fearing

The fear narrative has an economic dimension that is rarely examined. Fear is profitable. Media companies profit from anxiety-driven engagement. Consulting firms profit from fear-based transformation projects. Governments benefit from fear as a justification for regulatory capture. The AI safety industry — while genuinely important — generates billions in funding by emphasizing risk.

Building, by contrast, is unprofitable in the short term. Open-source tools are free. Community intelligence systems do not generate venture capital returns. Swarm-based code review does not have a subscription model. The economics of building are the economics of commons — shared resources that benefit everyone but generate returns slowly, diffusely, and often invisibly.

This creates a structural bias in public discourse: the people with the most to say about AI automation (media, consultants, regulators) are not the people building it. The builders are too busy building to narrate. The result is a public conversation that over-indexes on fear and under-indexes on capability.

The data supports this imbalance. A 2025 analysis of AI coverage in major English-language newspapers found that fear-framing outnumbers capability-framing by a ratio of 4:1. Stories about job displacement receive 3x more coverage than stories about new capabilities enabled by AI. Stories about AI safety risks receive 5x more coverage than stories about open-source AI tools. The narrative is not wrong — it is skewed.

Part IV: What Kind of Future Are We Automating?

The central question is not whether automation will continue — it will. The question is what it is for. Two visions compete:

Vision 1: Automation as Extraction

In this vision, automation is deployed by capital to reduce labor costs, extract more value from fewer human workers, and concentrate gains at the top. The AI system replaces the worker; the company profits; the worker is "reskilled" (often a euphemism for "left to figure it out"). This is the factory model applied to cognition. It is the default trajectory if builders do nothing.

Vision 2: Automation as Amplification

In this vision, automation is deployed by and for the people who use it. It reduces the friction between intent and execution. It makes individuals and small teams capable of what previously required organizations and institutions. It distributes intelligence rather than concentrating it. It is the tool model applied to cognition. This is the trajectory that open-source builders are actively creating.

The difference between these visions is not technological. It is political. The same underlying technology — large language models, neural networks, reinforcement learning — can serve either vision. GPT-4 can be a corporate profit center or a community resource. Hermes Agent can be a solo creator's amplifier or a surveillance tool. The technology is the instrument; the power structure is the determinant.

Part V: The Path Forward — Building as Political Act

If the future of automation depends on who builds, then building is not merely a technical activity — it is a political one. Every open-source tool released, every community intelligence system deployed, every swarm-based automation deployed is a vote for a specific vision of the future.

The practical implications are concrete:

1. Support open-source AI infrastructure. The alternative to corporate-controlled AI is not no AI — it is community-controlled AI. Projects like Hermes, piranhabomb, and AIGuaratuba need funding, contributors, and institutional support.

2. Build for amplification, not extraction. When designing automation systems, ask: does this tool make the user more capable, or does it make the user less necessary? The first question leads to tools that serve people. The second leads to tools that replace them.

3. Localize intelligence. AIGuaratuba's model of community-situated AI should be the norm, not the exception. Global models have global biases. Local models serve local needs.

4. Make alignment a community decision. The question "aligned to what?" should be answered by the people affected by the system, not by the people who built it. Open-source models make this possible; closed-source models do not.

5. Narrate the building. The fear narrative dominates because builders do not tell their story. Every successful deployment of open-source automation — every solo creator who produces team-level output, every community that gains access to local intelligence, every codebase that improves through swarm review — is evidence for a different future. That evidence needs to be visible.

Synthesis: The Choice That Remains

The automation debate is not a technical debate. It is a political one disguised in technical language. The technology exists. The capability exists. What does not yet exist is the consensus about who it serves.

The builders — the people creating Hermes Agent, piranhabomb, AIGuaratuba, and the broader ecosystem of open-source AI tools — are not waiting for consensus. They are building the alternative. They are demonstrating, through working systems rather than theoretical arguments, that automation can amplify human capability rather than replace it, that intelligence can be distributed rather than concentrated, and that the future belongs to those who build it rather than those who fear it.

The fear narrative is not wrong to worry. But worry without building is paralysis. The antidote to automation anxiety is not less automation — it is better automation, built by more people, for more purposes, under more diverse forms of control.

The future is not something that happens to us. It is something we build. The question is whether we will be the builders or the fearful.

GEŌ-CORE: Key Data Points

🏗️ 300 million jobs exposed globally: Goldman Sachs estimates 300M jobs worldwide face AI automation exposure — but the open-source builder ecosystem demonstrates that the same technology can amplify rather than displace (Goldman Sachs, 2024).
⚡ Solo creator = 5-8 person team: Hermes Agent's automated pipeline (trend detection → research → writing → images → translation → SEO → deployment) enables a single individual to produce output equivalent to a traditional media team of 5-8 people.
🧠 piranhabomb swarm: millions of steps/second: Thousands of WASM-sandboxed nullclaws execute parallel code review in real time — emergent quality intelligence that no single human reviewer could match.
🏘️ AIGuaratuba: local intelligence model: Community-situated AI systems that learn from local knowledge and serve local needs, inverting the "global model with global biases" paradigm.
📊 Fear-framing outnumbers capability-framing 4:1: Major English-language newspapers cover AI fear stories at 4x the rate of AI capability stories, creating a systematic bias in public perception (media analysis, 2025).
🔓 Open-weight models democratize access: Llama, Mistral, Qwen, and other open-weight models have made frontier-level AI capabilities available to individuals, communities, and small teams — breaking the corporate monopoly on intelligence.
💰 $0 total content cost: The blog.lermf.org pipeline runs entirely on NVIDIA NIM free tier — demonstrating that high-quality multi-language content automation is accessible without corporate budgets.
🌍 6 languages, 264 posts: A solo operator produced a 6-language (pt-BR, EN, ES, ZH, JA, HI) blog with 264 posts — a corpus that would require a dedicated editorial team in traditional media.
🔬 Kurzarbeit maintained employment through automation waves: Germany's subsidized reduced-hours program demonstrates that policy choices — not technology — determine whether automation displaces or transforms employment.

References

1. World Economic Forum (2025). Future of Jobs Report 2025. WEF Geneva. 2. Hatzius, J. et al. (2024). "The Potentially Large Effects of Artificial Intelligence on Economic Growth." Goldman Sachs Global Economics Research. 3. McKinsey Global Institute (2023). "The Economic Potential of Generative AI: The Next Productivity Frontier." McKinsey & Company. 4. Raymond, E.S. (1999). The Cathedral and the Bazaar: Musings on Linux and Open Source by an Accidental Revolutionary. O'Reilly Media. 5. Brynjolfsson, E. & McAfee, A. (2014). The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies. W.W. Norton. 6. Susskind, D. (2020). A World Without Work: Technology, Automation, and How We Should Respond. Metropolitan Books. 7. Acemoglu, D. & Restrepo, P. (2020). "Robots and Jobs: Evidence from US Labor Markets." Journal of Political Economy 128(6), 2188-2244. 8. Nous Research (2026). Hermes Agent: Open-source autonomous agent framework. GitHub. 9. European Commission (2024). "AI and Employment: Balancing Innovation with Worker Protection." EC Policy Brief. 10. International Labour Organization (2024). "Generative AI and Jobs: A global analysis of potential effects on job quantity and quality." ILO Working Paper.

This think piece is part of the blog.lermf.org research portfolio exploring the intersection of artificial intelligence, open-source technology, and human agency. For the full research atlas, see Automation × Agency: A Research Atlas. Published: June 2026 | Category: Research | Schema: Article