Most enterprises have deployed AI chat—fewer than 5% have agents truly in production
Getting Claude and Codex to employees is simple—90% of enterprises have done it. Actually putting agentic systems into production for non-engineering tasks, only 5% have managed it. That gap between deployment and real use is the episode's core data.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Story points as pay: top-to-bottom span is just 4×
Tenex pays engineers in 'story points' rather than hourly rates. Founder Alex's bet is that AI will create unprecedented variance in individual output—but currently the gap between top and bottom earners is only about 4×, far from the 10× dream. The real difficulty isn't detecting engineers who inflate story points—that's checked by an AI strategist role paid on customer retention and growth. The hard problem is pricing the hours engineers spend on architecture and customer communication, not writing code.
Opus 4.5 to 5.5 in one year has doubled per-engineer output
CJ sees last December's Opus 4.5 as the breakpoint where models could finally handle long-running tasks. By now, with Opus 5.5, the same engineers complete roughly twice as much work in the same time. But the story-point yardstick itself is loosening—the company recalibrates its story-point standards at least quarterly, because as models improve, each story point needs to deflate in dollar terms, or the company ends up paying twice for the same productivity gain.
— CJToken spend has climbed to 2× the cost of headcount
Most AI startups spend roughly as much on tokens as on headcount—the high end is 5 to 10×. Tenex reports 1:1 to 2:1, above the average. They did worry about token-maxing becoming the cost bottleneck, but they judge that Opus 5.5 has pushed the model to a Pareto point in cost-per-task—six to eight months ago was the real window to worry. Now models are dropping price faster than usage grows; most economically valuable tasks run fine on Opus 5.5, and it's not worth paying for a 10× smarter model.
— AlexEnterprise agent adoption is a collapsing funnel at every step
Deploying Claude or Codex to employees: 90 out of 100 enterprises do it. Training and enablement: drops to 60%. Agentic systems for engineering teams: most people estimate 70–80%, but it's really only about 50%. Actually running agentic systems in production for non-engineering work: 5%. That final leap is the widest chasm between FOMO-driven surface adoption and genuinely delegating workflow to agents.
— AlexVertical AI companies like Harvey are structurally doomed despite current success
Facing Harvey and OpenEvidence—vertical-AI companies that have won their niches—Alex and CJ are bearish overall, even though these firms are winning today. The reason: enterprise customers want bespoke 1:1 experiences. That forces software vendors to build out full FDE teams to sit with customers; software and services are converging, and software margins are disappearing. The only moat left is data that ChatGPT can't access—and both Anthropic and OpenAI are acquiring that, they're just not as aggressive about it as vertical players.
— AlexA $40K annual SaaS contract replaced with $1K of self-built AI
In Alex's 'Kill My SaaS' competition, a mid-tier SaaS quoted at $40K/year was replaced with a self-built AI alternative for $1K, ready to deploy at the next customer meeting. But he stresses that the higher you go in the stack—Salesforce, SAP, systems of record—the worse a self-build pitch becomes. One mistake carries a bet of hundreds of millions to billions, so enterprises naturally avoid that risk. He rejects claims that Salesforce or SAP face disruption.
— AlexAI replaces jobs, not people—but that's still a real hole
CJ notes a16z data showing call centers actually shrinking as AI takes over such work. If that's a skill someone built a career on, AI displacement is catastrophic. But when pitching agent deployments to customers, the core pitch is always 'reallocation,' not 'layoff': freeing people from the deepest technical details to cross-team collaboration and bigger-picture work. That loop of 'next problem' almost never stops in large organizations. Alex adds that employees should ask themselves how replaceable their role will be in a few years, not treat upskilling solely as the employer's responsibility.
New York's AI circle is smaller but its commercial pool may be larger
Alex describes New York's AI bubble as smaller than San Francisco's, but Tenex and their customers are themselves the 'New York version of San Francisco.' His thesis is that New York doesn't need to copy SF's scale; it should pull SF people to New York to meet customers with real money—a 'reality check.' New York's edge isn't the size of its AI circle; it's that the sheer commercial volume here can let AI technology truly permeate GDP.
— AlexIn their own words · checked verbatim
it almost feels like we've kind of hit this like Pareto point of like costs are coming down at a per task basis. Then it is way harder today for someone to token max and like really F the company in terms of cost than it was six to eight months ago.
Alex15:11
most enterprises that we talk to that have, let's just say more than a thousand employees do not have an agent in production that is doing work that is not building software.
Alex19:29
I phrase it to the team, is like, you can't automate a process that does not exist.
Alex23:22
there is so much nuance and so much inertia and so many systems you have to wrangle and people you have to wrangle.
Alex24:42
But why do I pay the $50,000 a year for Bloomberg? Well, it's just like that's that's what everyone, all my peers do. And it's all consolidated for me and it's all specialized for me.
Alex24:42
we were quoted $40,000 a year. We spent $1,000 to replace it.
Alex30:32
I think A16Z posted something about how like call centers are truly now on the decline and AI is starting to handle more and more of that work. And ultimately, if that is a skill you've built up your whole career, that is devastating to have something come in and take that function over.
CJ35:23
Figures
| Tenex team size | 75 | 3:17 |
| Internal pay spread (engineer top to bottom) | ~4× | 3:17 |
| Story-point output growth (year-over-year) | ~2× | 10:25 |
| Tenex token-to-headcount cost ratio | 1:1 to 2:1 | 15:11 |
| Industry average token-to-headcount ratio (AI startups) | ~1× (high: 5–10×) | 15:11 |
| Enterprises deployed LLM to employees | 90% | 19:29 |
| Completed AI training and enablement | 60% | 19:29 |
| Deployed agentic systems for engineering teams | ~50% | 19:29 |
| Deployed agents in non-engineering production | 5% | 19:29 |
Glossary
- MetaHarness
- Tenex's standardized documentation system and CLI tools for agentic workflows, unifying how to write plans, log work, and deliver output.
- FDE (forward-deployed engineer)
- An engineer stationed on-site with customers to provide deep-custom delivery; a key hiring focus for Tenex.
- RBAC (role-based access control)
- The permission framework determining which team members can access which data via agentic systems.
- AMP
- CJ's daily multi-model agentic programming tool that can switch between models while driving the same task.
- Kill My SaaS
- Alex's initiative to replace mid-market SaaS contracts with AI self-build alternatives.
How to listen
Enterprise AI transformation consultants, B2B product builders, or digital officers who want to know which step in the adoption funnel is actually blocking enterprises.
Opening chat and self-introductions (0:00-1:32) are low-information and can be skipped.