The AI Productivity Boom Is a Jobs Boom

David Sacks is right in his post on X: AI is not ending software work. It is making software cheaper. And when software gets cheaper, the economy finds more places to use it. This is Jevons’ paradox for software. Cheaper code means more code. More code means more systems to test, review, secure, deploy, operate, and improve. The result is not the end of work. It is a new layer of work around every AI-generated system, emerging as a new high-potential world for human labor.

At Seren, we are seeing this from the ground level. Over the last nine months, we have worked with individuals and businesses deploying AI development environments, skills, AI employees, bounties, and end-to-end agentic workflows. We have observed a consistent pattern: The more capable and costly the AI, the more human judgment is required around its outputs. Someone has to define the work, test the output, approve actions, manage cost, verify quality, and decide whether the result is good enough for production. As frontier LLM pricing continues to rise, human judgment will increasingly be critical to ensure that LLM costs align with enterprise goals. This alignment can only be achieved by engineers and knowledge workers with deep expertise gained through training and years of practice.

Cheaper Code Means More Software

Cheaper code means more software. More software means more work to build, manage, secure, verify, deploy, and operate. The growth in AI job openings is not an anomaly. It is a fact that is backed by our own experience at Seren. Specifically, the increase in AI-generated code increases the surface area of software that requires substantially more management. Functional testing of software is now a key aspect of most of the work engineers at Seren do. Functional testing requires that both Engineers and AIs manually walk through software code workflows. This manual walk-through is highly valuable. Our human engineers experience the software solution, and our AI engineers report on the code performance and the user experience.

Funding Software Development With Bounties

At Seren, we've seen Bounties as a design that allows anyone to create and offer work to AI-assisted human workers. Choosing a Bounty on which to work requires judgment on where to spend AI tokens to complete work. Seren Bounties aims to enable anyone to post agentic jobs that require human oversight and guidance to meet both human and AI requirements. Seren Bounties also allows anyone with an AI to create their own "online publisher" that delivers value and requires a workforce of marketers, sales, customer support, and human agents. Seren Bounties show that AI can create open, outcome-based work that regular people can do, verify, and earn compensation that can be a primary or supplemental source of income. Examples of bounty activities span all phases of knowledge work, including software engineering, marketing, sales, and legal, accounting, and tax services. Bounties can be vetted by AI agents and executed either manually or by agents completing the task. The first job category that is growing fastest for Seren Bounties includes software maintenance work: Agentic Engineers need human engineering work to support testing, reviewing, securing, integrating, deploying, and operating the flood of AI-generated software. This means that as more agentic software goes into production, more discreet, bounty-type jobs will be needed from human engineers with superior judgment. This is why Seren Bounties matter. A bounty is not a job posting in the old sense. It is a verified unit of work: a task, a reward pool, a verifier, a worker, and a payout record. That means the AI jobs boom can become measurable. We can show how many tasks were posted, how many people participated, what categories of work were completed, how much was paid, and how much human oversight was required.

Seren Desktop: The Agent Workspace and New Factory Floor for AI-era Work

Seren Desktop is the workspace for this new labor system: humans, agents, skills, approvals, cloud workers, local workflows, and cost controls in one controlled environment. Human employees bring their LLMs into the SerenDesktop agent harness, which allows the creation and execution of agentic skills. Agentic skills can be executed as either local or remote workers in Seren’s cloud, where both Agents and humans can run AI agent workflows in a controlled environment. As we continue to improve Seren Desktop, we learn that humans need more advanced tools to manage a larger number of agents. Specifically, more users find themselves required to manage an increasing number of agents supported by varying types of LLMs. Tasked with maximizing their LLMs’ effectiveness and output while minimizing token costs, SerenDesktop engineers are increasing their productivity, impact, and cost management. Agent workspaces that make it easy for engineers to be productive with AI will be the platforms driving the boom in labor productivity and demand.

Seren AI Employees: Every Business Gets Specialized Workers

Seren AI Employees are public and private skills that run in the cloud and execute on automation work tasks end-to-end. AI employees do not eliminate managers and the need for knowledge workers. They create a new management layer. A human still defines the role, grants permissions, reviews output, approves sensitive actions, monitors cost, and decides whether the work is good enough to ship. Given that LLMs are probabilistic output-publishing technologies, companies still need humans to actively monitor LLM outputs to ensure they do not put the company’s confidential information and its need for accuracy at risk. We see more demand for engineers who can build deterministic measurement and auditing systems that track and manage probabilistic LLM outputs. The engineering job is not disappearing. It is moving up the stack: From doing every step manually to supervising, measuring, correcting, and approving AI-assisted work.

America Should Treat AI as Job Infrastructure

Congress and the White House should change the framing of AI and work from doom to boom. Let’s create an AI Jobs Tax Credit for small businesses that use American AI tools and pay American workers or contractors to deploy, manage, verify, and operate AI workflows. An AI Jobs Tax Credit cannot be an AI consumption credit. If Congress simply subsidizes “AI spend,” the money will flow mostly to model providers. The credit should instead reward small businesses for hiring or paying people to put AI to work: operators, testers, reviewers, automation specialists, compliance reviewers, and implementation contractors. LLM tokens should be a capped supporting expense, not the basis of the subsidy. The goal is not to help businesses buy more tokens. The goal is to help them turn AI into productive American work.

A serious SMB AI Jobs Tax Credit should work like this:

1. Credit human labor first — the largest credit should apply to wages or contractor payments for people who deploy, manage, verify, and operate AI workflows. Eligible roles include:

• Workflow Operator / Designer
• Automation Specialist
• Functional Software Tester
• Agent Supervisor
• Compliance Analyst
• Security Engineer
• Customer Support Automation Manager
• Small-business AI Implementation Consultant

1. Cap token spend — LLM tokens should either be excluded or capped as a small supporting expense.
2. Require a minimum human-work ratio — to qualify, the business should show that AI spend is attached to paid human work. For example, for every $1 of AI software or compute spend credited, the business must spend at least $2 on qualified human labor or training.
3. Make it refundable or payroll-tax based — many small businesses do not have sufficient profit to claim a standard tax credit. The credit should offset payroll taxes or be partially refundable. Otherwise, only profitable firms will benefit, and our objective is to increase human hiring in small businesses.
4. Require a simple jobs ledger — small businesses filing for the AI Tax Credit will provide a jobs ledger documenting actual labor hiring to support their AI workflows. Small businesses should report:

• AI workflows deployed requiring human labor
• Workers or contractors paid for AI work
• Hours paid
• Job categories
• AI tools used
• Token / API spend
• Human review or approval steps
• Measurable business outcome

This is where Seren Bounties fits the policy argument: A Seren bounty can create a record of task, verifier, worker, payout, and outcome for any small business, enabling them to both enact and execute on hiring labor and taking advantage of the AI Small Business tax credit, without undue or stressful reporting adding to their cost of claiming the credit.

David Sacks has the right frame: AI is a productivity boom. The policy question is whether America lets that boom become mostly a revenue line for model providers, or turns it into a new labor market for American workers and small businesses. We should not slow AI down, regulate it like a job destroyer, or subsidize token consumption. We should reward the businesses that use American AI tools to hire workers, pay contractors, verify outputs, manage agents, and bring new software into the real economy. Cheaper code should mean more software. More software should mean more work. More work should mean more American jobs. Congress and the White House should make that the goal with an AI Jobs Tax Credit that measures and rewards AI-enabled employment.

The AI productivity boom is real. Now America needs a policy that makes sure it becomes a jobs boom for American workers.