Tenex pays engineers a base salary plus quarterly compensation tied to the story points they complete.
2
The model is intended to give engineers a direct incentive to use AI tools while preserving quality through strategist estimates, internal QA, and client approval.
3
Arman Hezarkhani says the system has supported fast delivery, including a billboard moderation model built in two weeks with 96% accuracy against human moderators.
Summary
Arman Hezarkhani explains Tenex's compensation model for AI engineering work. Clients are charged for delivered story points, and engineers receive a flat base salary plus a quarterly payment based on the story points they complete. He argues that hourly pay discourages speed, while salary, bonuses, and equity do not give every engineer a direct reason to adopt AI tools. Tenex pairs the payment system with safeguards. Strategists scope tickets, engineers write architecture documents and code, and each ticket goes through internal QA and client approval. He describes a billboard moderation system delivered in two weeks at 96% accuracy compared with human moderation, along with several edge AI models for retail devices. He is open about the risks, including inflated estimates, rushed work, and aggressive internal competition. His answer is careful hiring and counterbalancing engineer incentives with strategist compensation based on customer happiness.
Tenex ties engineering pay to delivered story points
Hezarkhani says Tenex charges clients for the number of story points it delivers and pays engineers for the story points they complete. Engineers still receive a flat base salary, then receive an additional amount each quarter based on completed points. He presents this as a way to connect compensation with shipped work rather than time spent. The model applies to custom builds that follow a roadmap, product requirements, architecture design, and tickets. He does not ask the audience to adopt it. He wants them to consider how AI may change engineering compensation.
Hourly pay gives engineers little reason to finish faster
Hezarkhani's illustrative history starts with hourly work. A buyer estimates that a task will take five hours and offers $500 at $100 per hour. The engineer keeps the same amount if the task finishes early, while a slow estimate protects the engineer from downside. He says software engineers are poor at estimating duration, so they may quote 15 or 20 hours instead. Salary and bonuses create a different problem in his example: employees can work a standard schedule without a direct connection to output. Equity helped startups create upside, but many companies fail and many engineers prefer cash.
AI changes the incentive to learn and use engineering tools
Hezarkhani compares his own workstation, where he had 45 agents handling tasks such as lunch orders, coding, and research, with a nearby engineer typing individual characters into Visual Studio Code without a chat window. He attributes the difference to incentives. As a founder, he wanted every increment of value and spent time following new software engineering tools. A salaried engineer with an annual bonus and equity may have less reason to search for tools that increase output. He argues that compensation should directly encourage effective AI use while retaining high code quality.
Strategists and engineers divide the work and balance incentives
Tenex has two client-facing roles. Strategists, who include former product managers and engineers, refine requirements with clients. AI engineers turn those requirements into architecture design documents, tickets, and code. Most engineering time goes into the design document, according to Hezarkhani. Each ticket receives a story-point estimate, and the engineer is paid when the ticket is accepted. Strategists are compensated based on NR, which Hezarkhani describes as being like customer happiness. Their incentives counterbalance the engineer's output-based payment because strategists scope the work and participate in QA.
A billboard moderation project connected fast delivery to a business cost
A billboard company had two rounds of image moderation, one inside the company and one with the billboard owner. Human moderation cost money and delayed the point when an advertisement could run, which also reduced revenue. Tenex scoped an AI moderation model, created an architecture design document, broke the work into tickets, and built it in two weeks. Hezarkhani reports 96% accuracy compared with the human moderator. The example shows how Tenex applies its story-point process to a concrete product problem where delivery time affects the client's operations.
Edge AI work uses parallel models on low-power retail devices
Another client had low-power devices in retail locations. One on-device model already generated heat maps from camera data, showing traffic patterns through a store. Tenex explored what additional capabilities could run if the hardware had more power and the models were quantized for parallel execution. The team built five models that could run in parallel. Hezarkhani names heat mapping, queue detection, and theft detection among their functions. He describes the same sequence as the other projects: start with product requirements, turn them into architecture, build the system, and pay engineers based on output.
The model needs controls against inflated estimates and rushed work
Hezarkhani names several risks. An engineer could inflate a ticket by assigning 45 story points to a simple button. An engineer could also rush through work because payment is tied to completion, causing quality to fall. He worries about creating a sharp-elbowed culture modeled too closely on sales competition. Tenex responds by having strategists scope the tickets, reviewing the work, and requiring multiple rounds of internal QA. Clients also approve every delivered ticket. These checks make the story-point count and the acceptance decision harder for one person to control.
Hiring quality matters more when AI amplifies individual habits
Hezarkhani says Tenex makes hiring very difficult so that other parts of the company become easier. His co-founder Alex compares AI to a distorted mirror that magnifies a person's attributes. In Hezarkhani's version, a strong engineer becomes better with AI, while a weak engineer becomes sloppier. He therefore treats hiring as the first control on the compensation model. He also argues that existing compensation systems can hold people back from using AI effectively, and asks listeners to consider different ways to reward potential and output in their own teams.
"There's no reason to work faster, right? And in fact, there's a disincentive to work faster."05:11
Who should watch
Engineering leaders considering compensation systems that reward shipped work rather than time spent will get Tenex's design and its internal checks.
Teams introducing AI coding tools may find the argument about incentives useful if adoption is uneven across engineers.
Anyone evaluating output-based pay should pay attention to the risks Hezarkhani names, especially estimate inflation, rushed quality, and sales-like competition.
