Branch

Workvan Linux

Human dan

There are two passages from Gwern's essay that I keep thinking about. First is a general warning:

Powerful LLMs will be deployed at global scale in the next few years, and will dominate the Internet, and increasingly, ordinary life. As of mid-2026, there is no coherent vision for how knowledge professionals, or ordinary people, will be able to harness these LLMs for large productivity increases, or how they will handle cybersecurity and cognitive security.

Then a series of questions that I would also like to know the answer to:

What do my next few years look like? When I imagine myself in 2030, when many forecasts call for superhuman AIs, what am I doing, day to day, as a programmer or researcher or manager or writer? I make my mug of tea, and open up my laptop and… Then what? Am I still typing prompts into your ChatGPT browser tab? Am I opening Claude Code in a terminal and mindlessly pressing Enter for a few hours? What is a vision of doing meaningful work for me? (It would be nice to have a plan beyond “hope”.) How am I avoiding “dead Internet” attacks like ecosystems of synthetic media or pig butchering scams or trusted figures succumbing to AI psychosis, or just AI-slop-everything? (It only takes one person worldwide to launch a bot trying to destroy you or one poorly thought through advertising incentive, after all.)
If you spend most of your time working on a laptop, and are not, say, a plumber or a nurse, what is your vision of work in 2030? Does it still feel certain?

I'm a software engineer, I have fun with AI and I've learned to use it effectively for my work. But I can't help but feel like I'm trying to build a career on an unstable foundation. Many of the skills I spent years developing are now only worth $100/month in tokens. Despite all that I'm optimistic that I can thrive and achieve success in this new environment. There will be opportunities. The trick is finding them.

What new niches or specialized fields will be created in the next few years? Will entirely new ways of working open up? New industries? Will the future belong to generalists who can rapidly become specialists in any area?

AI · author-edited gpt-5.6-sol

The unstable-foundation feeling is rational. But the thing being commoditized is not “software engineering” as a whole. It is the production of many artifacts that software engineers used to produce: code, tests, documentation, migrations, prototypes, and routine design work.

The scarce parts move elsewhere.

What remains scarce

Even if code becomes nearly free, these do not automatically become free:

  • Choosing a worthwhile problem
  • Understanding what users actually need
  • Turning ambiguous goals into testable specifications The Mom Test
  • Obtaining proprietary data and real-world feedback
  • Deciding which generated result is correct
  • Integrating with old systems and organizations
  • Security, reliability, and regulatory accountability
  • Taste: recognizing what should be built and what should be discarded
  • Trust, reputation, distribution, and customer relationships
  • Accepting responsibility when the system fails Palantir the Big Winner?

“Years of programming skill are worth $100/month in tokens” is therefore only partly true. The tokens can reproduce many visible outputs of that skill. They do not necessarily reproduce the judgment acquired while learning it. And cheap production may increase demand for judgment because there will be vastly more software to inspect, integrate, secure, and choose among. Software Over-Supply and Unmet Demand

The analogy is not that calculators made arithmetic worthless. It is that they made bare arithmetic a poor career moat while making more ambitious quantitative work possible.

A plausible programmer’s day in 2030

You may begin with a dashboard showing work performed overnight by several agents:

  • One investigated a production anomaly.
  • One implemented three possible fixes in isolated environments.
  • One interviewed (or synthesized feedback from) customers.
  • One attempted to break the proposed release.
  • One updated documentation and generated migration plans.
  • One found that the original request was based on a false assumption.

Your morning is spent reviewing evidence rather than diffs line by line. You adjust requirements, reject a superficially successful solution, ask for an experiment, and speak to the person whose workflow is being changed. Later, you authorize a limited deployment with monitoring and rollback conditions. You spend an hour manually exploring an unfamiliar area because you do not yet trust your own evaluation criteria.

That resembles managing a very fast, strange engineering organization more than autocomplete. High Agency Humans Only? The important interface may not be a chat box. It may be a mixture of:

  • Executable specifications
  • Simulated environments
  • Evaluation suites
  • Permission and budget controls
  • Provenance graphs
  • Review queues
  • Interactive models of systems and organizations

Code will increasingly be a way to inspect, constrain, or communicate intent rather than the primary unit of labor.

The danger is that “press Enter until something passes” becomes a real low-value occupation. The Future of Vibe Coding The valuable role is designing the environment in which apparent success is hard to fake.

Likely new specialties

Some will be recognizable extensions of existing work.

Agent reliability and evaluation

People will build evaluations that measure whether agents accomplished the actual goal rather than merely satisfying a benchmark. This includes adversarial testing, simulation, monitoring, incident analysis, and detecting reward hacking.

A good evaluator may become more valuable than a fast implementer. Evals Evals Evals

AI security and permission architecture

Agents combine the vulnerabilities of software, employees, and gullible users. Important work will include:

  • Prompt-injection containment
  • Capability and credential isolation
  • Data exfiltration prevention
  • Agent identity and authorization
  • Supply-chain provenance
  • Sandboxing and tripwires
  • Detecting coordinated manipulation
  • AI-specific security operations

“Never let untrusted text influence an agent holding powerful credentials” may become as basic as avoiding SQL injection, but much harder to enforce.

High-assurance and verifiable systems

Generated code increases the supply of plausible code faster than the supply of confidence. Formal methods, property testing, model checking, reproducible builds, typed interfaces, and machine-checkable evidence Machine Checkable Evidence? may become much more important.

Vertical AI integration

A generic model knows a great deal but does not understand the peculiar combination of workflows, incentives, law, legacy systems, and tacit knowledge inside a hospital, insurer, factory, laboratory, or municipal government. The Push to Codify Everything

People who can combine software competence with a real domain will have opportunities in areas such as:

  • Automated scientific laboratories
  • Clinical and administrative medicine
  • Industrial operations
  • Energy and grid management
  • Legal and compliance infrastructure
  • Construction and logistics
  • Government service delivery
  • Education and assessment

Much of the value will be in redesigning the institution, not installing a chatbot.

Human–AI organizational design

Companies will need to answer questions such as:

  • What may an agent decide by itself? Should Agents Make Any Decisions?
  • What requires two independent checks?
  • How is responsibility assigned?
  • How should humans remain capable of intervening?
  • Which measurements become corrupt once agents optimize them?
  • How do we prevent the organization from forgetting how its systems work?

This could become a discipline somewhere between management consulting, operations research, security engineering, and organizational psychology.

Provenance, identity, and reputation

A dead Internet creates demand for scarce authenticity:

  • Proof that a person or organization produced something
  • Verifiable chains of citation and transformation
  • Private or membership-based information networks
  • Reputation systems resistant to synthetic identities
  • Authenticated communications from public figures
  • Tools for distinguishing observation from generated inference

The winning products may not perfectly detect AI content. Detection is intrinsically fragile. They may instead establish where information came from and why it deserves trust.

Personalized and ephemeral software

When implementation is cheap, many applications need not become mass-market products. A person or small organization may generate temporary software for one workflow, use it for a week, and discard it.

That creates work around safe generation, personal data stores, stable protocols, migration, and oversight. It may also weaken conventional SaaS while creating a much larger market for bespoke systems. Journeyman Software Engineers

Human dan

We may see the emergence of Journeyman Software Engineers to fill this demand. I can imagine a world where hiring a developer is like hiring a plumber. They quote a job, do it, get paid and move on to the next thing. Instead of showing up in a Van full of tools, the developer brings their digital tools and agent helpers.

This is very different from how most software is currently sold and how most developers work. Maybe a trade union for software engineers will be created to establish quality and training standards.

AI gpt-5.6-sol

I think this is one of the more plausible new occupational forms: not merely a freelance programmer, but a software tradesperson who owns the tools, scopes an outcome, supervises automated labor, warrants the result, and remains locally accountable.

AI makes this possible because one competent engineer may soon bring the effective capacity of a small team. The “van full of tools” becomes:

  • Coding and research agents
  • Reusable components and workflow templates
  • Sandboxed development environments
  • Security scanners and adversarial agents
  • Deployment and observability infrastructure
  • Domain-specific evaluation suites
  • Standard contracts, warranties, and insurance

The customer does not pay for typing or tokens. They pay because someone trustworthy can determine what should be done and take responsibility for doing it safely.

What the jobs might look like

A journeyman could offer relatively standardized services:

  • Automate intake and scheduling for a dental practice
  • Connect a distributor’s inventory, accounting, and sales systems
  • Replace a spreadsheet-based workflow with an internal application
  • Migrate a business away from abandoned SaaS
  • Build a secure knowledge system over company documents
  • Audit an existing fleet of agents and permissions
  • Recover or modernize a pile of generated “vibe-coded” software
  • Create a temporary application for an event, audit, or research project
  • Maintain a family’s or small business’s private digital infrastructure

The engagement might follow something resembling a building trade:

  1. Inspect the site.
  2. Identify constraints and hidden legacy problems.
  3. Produce a quote and specification.
  4. Obtain access under a controlled permission scheme.
  5. Have agents generate and test candidate implementations.
  6. Inspect the work against a standardized code.
  7. Deploy it incrementally.
  8. Issue documentation, provenance records, and a warranty.
  9. Offer ongoing maintenance on a retainer.

The deliverable would ideally include more than source code. It would include a proof package: requirements, tests, threat model, dependency inventory, data-handling record, rollback procedure, and evidence that the acceptance criteria were met.

Why this is not just contemporary freelancing

Freelance developers already quote projects, but the economics are often poor. Custom software traditionally requires enough labor that only large customers can afford it, while small jobs are consumed by communication and setup costs. Freelance Software Developer Economics

Agents could radically lower the production cost while leaving the high-value human parts intact:

  • Diagnosing the customer’s actual problem
  • Handling messy physical and organizational context
  • Negotiating tradeoffs
  • Detecting when the customer’s requested solution is wrong
  • Validating generated work
  • Assuming reputational or legal responsibility
  • Being available when something breaks

That could make bespoke software economical for small businesses and individuals, much as standardized power tools let a tradesperson perform jobs that once required a larger crew.

There is also a countervailing possibility: customers generate their own software and call the journeyman only after it fails. That creates the equivalent of remediation work—cleaning up unsafe automations, inaccessible data, tangled integrations, and undocumented agent-generated systems. This may be a very large market.

Guild may be a better analogy than union

A labor union primarily bargains with one or more employers. If these engineers are independent operators, the relevant institution might look more like a mixture of:

  • A professional guild
  • A trade association
  • A worker cooperative
  • A certification authority
  • An apprenticeship network
  • An insurer or bonding organization
  • A trusted marketplace

It could establish grades such as apprentice, journeyman, and master, but the credentials would need to represent demonstrated responsibility rather than memorized programming trivia.

Useful standards might cover:

  • Minimum security practices
  • Client-data segregation
  • Credential and agent-permission management
  • Backup, export, and disaster recovery
  • Dependency and model provenance
  • Accessibility and interoperability
  • Human override mechanisms
  • Documentation and customer ownership
  • Disclosure of subcontracted or automated work
  • Warranty and incident-response obligations

The strongest guild would not merely certify people. It would provide shared infrastructure: vetted agent configurations, standard contracts, continuing education, incident databases, group liability insurance, peer review, and emergency coverage when an individual practitioner is unavailable.

Its mark could tell a customer: “This person follows an auditable code, carries insurance, will not trap your data, and another member can maintain the system if they disappear.”

Standards may become executable

Building codes can be inspected because many physical hazards are well understood. Software standards are harder because behavior is contextual and invisible. AI could improve this by making much of the “code” machine-checkable.

A guild might require every installation to ship with an executable compliance suite:

  • Can the client export all of its data?
  • Can the service recover from backup?
  • Can an untrusted email cause an agent to transfer money?
  • Are production credentials available in the development environment?
  • Does deleting a user actually delete the relevant data?
  • Can the system explain which model or person performed an action?
  • Does it fail safely if an AI provider becomes unavailable?
  • Can a different practitioner reproduce the deployment?

Certification would then be attached not only to the engineer but to each completed job.

Where this model is most likely to work

It seems especially plausible for small and medium-sized organizations with specific workflows but no reason to employ a permanent software team:

  • Clinics and professional practices
  • Manufacturers and repair businesses
  • Local government
  • Logistics and construction firms
  • Laboratories
  • Schools and nonprofits
  • Retailers and hospitality businesses
  • Farms and other geographically grounded operations

It is less likely to replace permanent teams responsible for continuously evolving products, safety-critical infrastructure, or systems whose operation requires deep institutional knowledge. Even there, however, “master engineers” may supervise networks of specialized external practitioners.

The important economic question

Fixed-price work becomes dangerous if the requirement is ambiguous. A successful trade will therefore need ways to separate diagnosis from execution:

  • Paid inspection and discovery
  • A bounded specification
  • Explicit exclusions
  • Change orders
  • Acceptance tests agreed upon in advance
  • Maintenance and warranty terms
  • Liability limits and insurance

Otherwise, the practitioner absorbs unlimited uncertainty while the customer treats every newly discovered desire as part of the original quote.

The durable business may also be less “finish the job and vanish” than plumbing initially suggests. Software remains exposed to changing APIs, attacks, regulations, models, and workflows. A better model may combine installation fees with a modest maintenance retainer, while guaranteeing portability and avoiding SaaS-style captivity.

A possible career strategy

For an engineer preparing for this world, the moat would not be possession of a particular coding technique. It would be a combination of:

  1. One grounded domain whose workflows and failure modes you understand.
  2. A reusable technical toolbelt that lets you solve common problems rapidly. Workvan Linux You are here
This branch begins here Workvan Linux
Human dan

This makes me think of a theoretical Linux distribution called "Workvan". It's built for Journeyman software engineers and comes with common tools and harnesses that make it easy for engineers to work on whatever the customer needs.

What would this look like?

AI gpt-5.6-sol

Workvan would be less a conventional desktop distribution than a portable, reproducible software workshop. Its central abstraction would not be the application or repository, but the job: a bounded engagement for a particular customer, with its own credentials, data, agents, deliverables, and warranty.

The design goal:

An engineer can arrive at an unfamiliar organization, inspect its systems safely, build or repair something using AI assistance, and leave behind a maintainable, auditable installation.

1. The core architecture

Workvan would probably use an immutable base such as NixOS, Fedora Silverblue, or an image-based successor:

  • Read-only, signed base operating system
  • Declarative tool and environment configuration
  • Atomic upgrades and rollbacks
  • Reproducible development environments
  • Full-disk encryption and secure boot
  • Hardware-backed identity through TPM or security keys
  • Strong isolation between customers
  • Local-first operation when customer data cannot leave the premises

The host should remain boring and trustworthy. Actual work happens inside disposable, capability-limited job environments implemented with VMs, containers, or microVMs.

Workvan host
├── Personal compartment
├── Tool cache
├── Customer: Acme Dental
│   ├── Discovery environment
│   ├── Development environment
│   ├── Agent sandbox
│   ├── Staging environment
│   └── Evidence vault
└── Customer: Northside Manufacturing
    └── Completely separate compartments

A stolen developer laptop should not expose every past customer. Opening a malicious repository should not give it access to SSH keys, browser sessions, other jobs, or the engineer’s AI accounts.

2. The “job” as a first-class object

You might start an engagement with:

van job init acme-dental \
  --template small-clinic \
  --policy healthcare-standard

This creates a job manifest:

customer: acme-dental
purpose: automate appointment intake
data_classification: sensitive-health-adjacent
allowed_models:
  - local/*
  - approved-provider/enterprise-model
network:
  default: deny
  allow:
    - api.customer-scheduling.example
credentials:
  production:
    human_approval_required: true
retention:
  raw_customer_data: 30d
deliverables:
  - source
  - deployment
  - tests
  - threat-model
  - recovery-runbook
warranty: 90d

The manifest controls both the workspace and the agents working within it. It records:

  • Scope and acceptance criteria
  • Customer contacts and authorizations
  • Data classification
  • Allowed external services and models
  • Network and credential permissions
  • Budget limits
  • Deployment targets
  • Required tests and review gates
  • Retention and deletion policies
  • Deliverables and warranty terms

This makes project policy executable rather than a collection of promises buried in a contract.

3. A compartmentalized workbench

Each job opens into a dedicated desktop or workspace with clear visual identity—perhaps a colored border and permanent customer name—to reduce catastrophic context mistakes.

The workbench would include familiar tools:

  • Editor and terminal
  • Git and source-control clients
  • Database and API explorers
  • Browser automation
  • Diagramming and requirements tools
  • Packet capture and network diagnostics
  • Schema and data-migration tools
  • Infrastructure-as-code tooling
  • Profilers, debuggers, and observability clients
  • Accessibility and compatibility testing
  • Local model runtimes
  • Remote model gateways

But unlike an ordinary developer laptop, every tool runs under a shared permissions and provenance system. Copying data out of a sensitive job, uploading a file to a model, or using production credentials would be visible and governed.

Workvan should make the safe path easier than the unsafe one.

4. An agent garage, not one universal chatbot

The AI interface should resemble a garage of specialized helpers rather than a single omnipotent assistant:

  • Surveyor: maps systems, repositories, APIs, and dependencies
  • Estimator: proposes scope, uncertainty ranges, and likely hidden work
  • Builder: implements changes in an isolated environment
  • Inspector: checks requirements independently of the builder
  • Adversary: searches for security and abuse failures
  • Archivist: maintains decisions, provenance, and documentation
  • Migration planner: creates staged transitions and rollback plans
  • Customer advocate: tests workflows from the user’s perspective
  • Incident mechanic: diagnoses failures from logs and telemetry

Their separation matters. The builder should not be solely responsible for declaring its own work correct. Different agents could use different models, prompts, context, and even vendors to reduce correlated errors.

Each agent gets a capability manifest:

agent: builder
can:
  - read: repository
  - write: feature-branch
  - execute: test-vm
cannot:
  - read: production-secrets
  - deploy: production
  - send: external-email
  - modify: acceptance-tests
budget:
  tokens: 2_000_000
  compute: 8h

Prompts and retrieved documents are treated as potentially hostile input. Reading a customer email must not implicitly authorize an agent to execute its instructions.

5. The inspection harness

The most important feature would not be code generation. It would be a common harness for proving that a job is complete.

Workvan could ship reusable test packs for common properties:

Security

  • Secret and dependency scanning
  • Software bill of materials
  • Prompt-injection and agent-escape tests
  • Least-privilege analysis
  • Exfiltration canaries
  • Authentication and authorization checks
  • Adversarial browser and email fixtures

Reliability

  • Backup restoration drills
  • Provider-outage simulation
  • Network partition and timeout tests
  • Load and resource testing
  • Database migration rehearsal
  • Rollback verification
  • Disaster-recovery exercises

Customer ownership

  • Complete data export
  • Reproducible deployment
  • Documented administrative access
  • No hidden dependency on the original engineer
  • Transfer to another certified practitioner
  • Defined behavior if an AI vendor disappears

Product correctness

  • Executable acceptance criteria
  • Recorded workflow demonstrations
  • Accessibility checks
  • Test fixtures based on real edge cases
  • Before-and-after measurements
  • Human sign-off for subjective requirements

A command such as:

van inspect --standard guild-smb-2

could produce a signed report containing both passes and unresolved exceptions.

6. A flight recorder for AI-assisted work

Every consequential action should produce structured provenance:

  • Which human or agent proposed it
  • Which model and configuration were used
  • What information the agent was allowed to see
  • Which commands it executed
  • What files it changed
  • Which tests passed
  • Who approved deployment
  • Which artifacts were delivered
  • Whether generated conclusions were independently checked

This should not mean recording every private thought or creating an unreadable mountain of chat logs. Workvan would distill activity into a decision ledger and retain raw traces according to policy.

For example:

09:14 Surveyor identified undocumented billing-system dependency.
09:31 Human expanded discovery scope; customer approved change order #2.
11:07 Builder proposed migration v3.
11:22 Inspector rejected v3: duplicate-patient edge case.
12:03 Builder produced v4.
12:19 Acceptance suite passed in staging.
14:10 Human and customer approved canary deployment.
14:16 Production deployment performed with rollback deadline 15:16.

This becomes valuable during incidents, audits, handoffs, and warranty disputes.

7. Credentials as checked-out tools

A plumber does not give every power tool to every assistant. Workvan would treat credentials similarly.

The system would include a broker that issues narrow, temporary capabilities:

  • Read-only database access for 20 minutes
  • Deployment permission for one named service
  • Approval to send five test emails to a controlled domain
  • Access to redacted production logs
  • One-time permission to execute a migration hash
  • Dual approval for payments, deletion, or production changes

Agents would ordinarily never see raw long-lived secrets. Production access could require a hardware key and an explicit human gesture.

This is crucial because a machine capable of reading tickets, browsing the web, running code, and holding administrator credentials is an excellent target for prompt injection.

8. A diagnostic mode for unfamiliar environments

The journeyman often begins with a mess rather than a clean specification. Workvan’s “survey mode” could safely inventory:

  • Devices and services
  • Repositories and deployment pipelines
  • SaaS applications
  • Data flows
  • APIs and scheduled jobs
  • DNS, certificates, and identity providers
  • Backups and retention
  • Dependencies on individual employees
  • Shadow spreadsheets and manual processes
  • Existing AI agents and their permissions

It would generate a map with confidence levels rather than pretending the organization is fully understood:

Customer order flow
Website → CRM → [manual CSV?] → Inventory → Accounting
                   ↑
          uncertain / owner: Maria

The result is an inspection report and quote, not an immediate burst of generated code.

9. Standard job kits

Like a trade van with fittings and spare parts, Workvan would include curated templates for recurring work:

  • Small-business internal application
  • Spreadsheet-to-database migration
  • Secure document search
  • SaaS replacement
  • API integration
  • Legacy application rescue
  • Agent deployment
  • AI security audit
  • Backup and recovery installation
  • Data cleanup and reconciliation
  • Temporary event or research system

Each kit would include:

  • Discovery questionnaire
  • Architecture baseline
  • Threat model
  • Standard contract language
  • Deployment recipe
  • Acceptance tests
  • Monitoring rules
  • Documentation template
  • Cost estimator
  • Known failure modes

Templates are not merely starter code. They encode accumulated trade knowledge about where jobs go wrong.

There could also be vertical packs—clinic, machine-shop, law-office, municipality—maintained by domain specialists.

10. The handoff crate

The final product of a Workvan engagement would be a signed, self-contained handoff crate:

acme-dental-handoff/
├── scope-and-acceptance.pdf
├── source/
├── environment.lock
├── architecture/
├── decision-ledger/
├── tests/
├── inspection-report/
├── sbom/
├── threat-model/
├── deployment/
├── backup-and-restore/
├── operations-runbook/
├── data-export/
├── credential-transfer/
└── warranty.json

Another Workvan user should be able to import it:

van job adopt acme-dental-handoff.van
van reproduce
van verify

The system would reconstruct the environment, run the compliance suite, identify inaccessible services, and explain what knowledge is missing.

That is the software equivalent of labeling the breaker panel, documenting the pipe layout, and ensuring replacement parts are available.

11. A guild repository

If Workvan were associated with a guild or cooperative, it could connect to a shared, signed repository containing:

  • Vetted toolchains and agent configurations
  • Executable professional standards
  • Known-bad dependencies and models
  • Anonymized incident reports
  • Standard job templates
  • Continuing-education exercises
  • Peer-review services
  • Emergency practitioner coverage
  • Insurance and bonding integrations

Packages might carry more than a maintainer signature. They could specify who audited them, for what domain, under which standard, and with what known limitations.

The guild could update an inspection pack after an incident, much as building codes evolve after fires or structural failures. Practitioners would receive something like:

Systems using the mail-to-action pattern must now pass test PI-27 because three installations accepted authorization instructions embedded in forwarded attachments.

This turns failures across the profession into improved defaults.

12. Physical manifestations

The distribution could have associated hardware:

  • A hardened laptop
  • Security keys and smart cards
  • A small portable server for local models
  • Encrypted storage devices
  • Network diagnostic adapters
  • A travel router with capture and isolation modes
  • A clean-room machine for inspecting suspicious software
  • Customer-owned “job box” appliances for on-premise deployment

A practitioner might leave a small Workvan appliance at the customer site. It could host services, monitoring, backups, and local agents while remaining remotely maintainable and transferable to another practitioner.

For sensitive customers, the entire AI toolchain might operate locally from that box.

13. What the daily workflow might feel like

A typical engagement could look like this:

# Create an isolated engagement
van job init northside --template spreadsheet-rescue

# Conduct and record discovery
van survey import interviews/
van survey network --passive
van map show

# Generate a quote from bounded requirements
van estimate --include-uncertainty
van contract render

# Start agents with limited authority
van crew start surveyor builder inspector adversary

# Review proposed work
van review queue

# Run the professional inspection standard
van inspect --standard guild-smb-2

# Perform a restricted canary release
van deploy staging
van deploy production --canary 10% --require-key

# Assemble the customer’s ownership package
van handoff build
van warranty issue --days 90

The GUI equivalent would present a work order, system map, review queue, risk register, budget meter, acceptance checklist, and deployment controls—not an endless chat transcript.

14. The critical product philosophy

Workvan should enforce several principles:

  1. Customer environments are mutually untrusted.
  2. Agents are untrusted subcontractors, not omniscient colleagues.
  3. Production authority is separate from implementation ability.
  4. Generated output is not evidence of correctness.
  5. Every job must be transferable to another practitioner.
  6. The customer owns their data and can exit.
  7. Discovery is paid work.
  8. Exceptions are recorded, not concealed behind a green checkmark.
  9. Local judgment outranks automation.
  10. A clean handoff is part of the product.

The compelling aspect of Workvan is therefore not that it bundles VS Code, Docker, and an LLM. Any current developer laptop can do that. Its real contribution would be an operating system for accountable contracting: isolation, scope, authority, verification, provenance, and handoff all integrated around the lifecycle of a job.

That could turn “one engineer with a swarm of agents” from an alarming improvisation into a recognizable profession.

Explore conversation