⚡️ 10x AI Engineers with $1m Salaries — Alex Lieberman & Arman Hezarkhani, Tenex

Arman shares how he was forced to downsize his previous company Parthian's engineering team by 90%, leading him to rebuild the entire product and engineering process AI-first. This resulted in 10x production-ready software output, which became the founding insight for Tenex's compensation model that pays engineers for output rather than hours.

• •Downsizing engineering team by 90% forced complete rearchitecture to AI-first development
• •Production-ready software output increased 10x after the shift
• •Traditional hourly billing perversely incentivizes slower work, even as AI enables unprecedented speed
• •Engineers leveraging AI can't charge proportionally higher hourly rates without being 'laughed out of the room'

The founders explain their story point-based compensation system and how they prevent gaming through strategic hiring and dual-role oversight. They hire 'long-term selfish' engineers who understand sustainable client relationships and pair AI engineers with technical strategists incentivized on retention.

• •Hire people who are 'long-term selfish' - understanding incentives extend beyond immediate story points
• •Technical strategists are incentivized on NRR and retention, acting as final quality gatekeepers
• •Engineers who genuinely enjoy coding and working with smart people naturally avoid gaming the system
• •No clients have yet argued about story point assignments despite being a young company

Tenex will have multiple engineers making over $1 million in cash compensation next year through story points alone. Projects include building computer vision systems for retail in two weeks, deploying a mobile app that hit #20 globally in one month, and using rapid prototyping to close sales deals within hours.

• •Multiple engineers projected to earn $1M+ annually based purely on story point compensation
• •Built retail computer vision system with multiple models on edge devices (Raspberry Pi, Jetson Nano) in 2 weeks - previously would take quarters
• •Snapback Sports mobile app built and deployed in one month, reached #20 globally on app store
• •Engineer built working fitness coach app prototype in 4 hours to close a sales deal that seemed dead

Tenex standardizes on TypeScript front-to-back with shared type schemas to maximize agent autonomy. The team doesn't have a 'favorite coding agent of the month' - they continuously evaluate which models perform best for specific tasks at any given moment, treating tool selection as a craft rather than a science.

• •Default stack: TypeScript front and back end with shared types/schemas folder for maximum structure
• •High structure enables agents to work autonomously for longer periods
• •Team tracks model performance hour-by-hour, switching between Claude, Codex, etc. based on current performance
• •Tool selection treated as 'feel' and craft - like a samurai's weapon - rather than purely eval-driven

Unlike many companies that dropped take-home interviews post-AI, Tenex uses 'unreasonably difficult' take-homes that 50% of candidates don't even attempt. This filters for exceptional engineers and keeps the overall interview process short - as quick as one week from first call to offer.

• •Take-home interviews are intentionally 'unreasonably difficult' - 50% of candidates don't respond
• •Entire interview process can be completed in one week: two calls, take-home, review, 1-2 final meetings
• •Key interview question: 'What would be the first major bottleneck to build a fully autonomous AI senior engineer?'
• •Best answer focused on 'controlling entropy' - even 1% error rates compound and derail autonomous agents

The team identifies context engineering - not just context length, but getting the right context to LLMs and making them pay attention to the right parts - as the fundamental bottleneck for autonomous AI engineers. Engineer Dan's insight about controlling entropy reveals how small error rates compound and derail agents.

• •Context engineering (not just length) is the key bottleneck: getting right context and attention mechanisms
• •Dan's entropy answer: even 99% accuracy means 1% error rate compounds and accumulates, derailing agents
• •Reducing entropy to achieve 100% accuracy per task prevents compounding errors in autonomous loops
• •This is the 'sophisticated version of context engineering' that experienced practitioners recognize

Tenex is currently 100% human capital constrained - finding and hiring enough exceptional engineers is what keeps the founders up at night. The second constraint is matching great people with the right processes to maintain delivery quality at scale.

• •Primary constraint: hiring enough good engineers fast enough
• •Secondary constraint: matching engineers with right processes to maintain delivery quality at scale
• •Technology is an enabler, not the bottleneck today
• •Long-term ambitions include building proprietary technology beyond consulting services

Arman controversially argues that MCP (Model Context Protocol) is essentially 'a three letter word for API' and criticizes the hype cycle around rebranding existing concepts. The discussion reveals how different groups in tech create new terminology to communicate, and the importance of learning through debate and contrast.

• •Arman's hot take: MCP is fundamentally just API wrappers with a new name
• •Criticism of using three-letter acronyms to raise money and create artificial hype
• •Some engineers have warmed Arman up to specific MCP use cases despite skepticism
• •Conference will host pro/con MCP debate to help people learn through contrast rather than agreement