# AI in NYC: How Tenex Builds an AI-Native Engineering Team — Alex Lieberman & CJ Hess

## Executive summary

Tenex discusses the shift in engineering from time-based labor to outcome-based intellectual output, driven by advanced AI agents. The firm emphasizes that AI requires engineers to build highly structured, 'AI-native' processes (like the MetaHarness) and that the biggest challenge in enterprise adoption is not the raw intelligence of the models, but the 'last mile' of diffusing that intelligence through existing corporate systems and processes. The model suggests that successful AI integration requires standardizing processes and focusing on agent deployment for non-engineering, back-office tasks.

## Key takeaways

- Shift to Outcome-Based Compensation: The future of engineering compensation is moving away from paying for time spent (hourly) toward rewarding measurable output (story points/outcomes). This model creates a wide dispersion in earning capacity, rewarding those who can best leverage AI tools.
- The MetaHarness Process: Tenex utilizes a structured workflow, the MetaHarness, which standardizes skills and processes. This involves logging all agent decisions, plans, and visualizations directly into Git, ensuring that the code and the context (the 'why') are always in a durable, accessible record.
- Enterprise AI Adoption Gap: While general LLMs are widely adopted (90% of surveyed enterprises have rolled out LLMs), the adoption of coding agents is only around 50%, and the deployment of agents for non-engineering, back-office work drops significantly lower (to 5%).
- The Value of Process Standardization: Agents cannot proactively automate a process that is not standardized. The primary recommendation for large companies is to first build a standardized process around the organization before attempting to hand off work to an agent.

## Technical details

- AI-Native SDLC and Context Management: Tenex's approach involves treating the entire development process—including plans, decisions, and visualizations—as code that is checked into Git. This ensures that the context and rationale for every decision are durable and accessible, giving agents 'superpowers' by centralizing knowledge.
- LLM Performance and Cost: The speaker noted a significant increase in engineer output, estimating a roughly 2x increase in capability since the beginning of the year. Regarding models, Opus 5.5 is currently favored for its balance of speed, cost, and intelligence. The speaker also highlighted the utility of using tools like Codeex for review and working back and forth on a local machine.
- Agent Deployment and Security: A major pain point in enterprise AI adoption is data permissioning. As every employee has access to LLMs (like Claude or Codeex), managing access to specific data sources becomes a 'nightmare' for CISOs, necessitating dedicated gateway solutions.

## Practical implications

- For engineering teams, focus on building standardized, documented processes (like a 'MetaHarness') rather than relying on ad-hoc knowledge transfer.
- When evaluating AI ROI, measure the value of process standardization and context capture (the 'why') alongside the code output.
- When planning AI adoption in large enterprises, prioritize building a structured process around the function *before* attempting to deploy an agent.
- Be aware that the most significant barrier to AI adoption is often governance and data permissioning, not model capability.

## Topics

AI Engineering, Software Development Lifecycle (SDLC), Agent Automation, Enterprise AI Adoption, Compensation Models, Tenex’s AI-native SDLC, Tenex AI engineering, Kill My SaaS results

Source: https://www.youtube.com/watch?v=Vki6cklFqxo