This is step 1 of a sweeping change that every enterprise is going to undergo in how tokens are distributed across employees. It starts with an arbitrary cap on token spend and will eventually move to a scorecard based system that tracks performance and multiple different heuristics to determine how much $$$ you can burn on tokens. It’s funny to see such a sharp negative reaction from twitter on this change including believing that innovation is going to quickly die at Accenture. I don’t disagree that a broad cap like this will negatively affect a few employees who know how to wield Claude super effectively but I’d be willing to bet you the *average* employee at Accenture is using AI in wildly inefficient ways that increase costs exponentially. So now the question is how quickly can Accenture distribute tokens to the right employees and build a system that scales that across 10k+ people. One of the more interesting enterprise challenges right now and one I’ve been seeing a ton of engineering teams struggle with even for coding agent consumption and rollout!

Yep, people are surprised to learn this but it’s the truth. And for many enterprises they are actually still struggling with rolling out coding agents. Why? Well it turns out adoption of AI is more of a change management problem at scale vs technical one. How do you roll out coding agents across 5k+ users in a way that doesn’t mess up one of cost, speed, or quality? Many enterprises are realizing they dropped millions on these tools and did NOT see the impact on metrics they were hoping. Does that mean coding agents === scam. No absolutely not. Any founder will tell you that their best engineers have become even more effective due to AI. The 10x Eng becomes 100x and I’ve seen it happen. But it turns out giving a coding agent to an engineer who was mediocre or an engineering team with weak foundation exposes it even faster. So now imagine at enterprise scale….. your PR throughput 5x but so did your MTTR and vulnerability across services. This is btw why you are seeing the labs invest big time in FDE models. They all realized that AI transformation is not unlike the on-prem to cloud movement that has taken more than a decade. It will take time but will eventually unlock billions in enterprise value

Today at @GetCortexApp we’re excited to announce that we’re open sourcing engrams (engramsfactory.com/) - our self hosted, software factory built for enterprise needs. It’s a bring-your-own-harness alternative to Devin (by @cognition ) and @FactoryAI - you get 80% of the value with 100% less of the drama ( iykyk 😂 ). What does engrams do and why should enterprises care? Engrams is exactly what you’d expect from an ADE / software factory platform like Devin, Factory, Ramp’s inspect, etc – except it’s fully open source. It’s the most powerful open source factory available in the market today. It has:  - long-lived, space-efficient, compute-efficient sandboxing/VMs for agents w/ credential brokering for secrets - embedded browser for human or agent use and preview environments - native IDE in every session - automations framework - slack integration - LLM gateway support - multi-player mode - harness-agnostic artifact sharing - so much more Enterprises care about cloud-agent software factories to manage risk, they want to avoid being locked into one lab like Anthropic or OpenAI. Engrams enables lets you be not just model agnostic, but *harness* agnostic as well – bring Claude Code, Codex, or any other harness you want! This seems too good to be true, why is Cortex open sourcing something that we’ve seen enterprises pay 7-8 figures for? Why not sell it as a product? Our belief is that in many cases the harness / ADE layer is going to be very quickly commoditized. Companies like Ramp are building their own. Both building and buying is leading to the exact same product capabilities. What we believe will be more important than ever is the AI governance and adoption layer. It’s easy to generate code but building an engineering org that can actually deploy and operate it in confidence is the hard part. It’s easy to “build an agent” but safely deploy it in your SDLC and ensuring it has the right context and governance to meet your standards is the hard part.  This is where the Cortex platform comes in – if you care about governance, quality, cost, and building a great engineering org, that’s when we make money. Not by selling tokens as a commodity. We hope to be a part of your journey with Engrams as you build your AI native SDLC. We announced this at Evolve and several of our larger enterprise customers are currently working on migrating from other software factories - which will save them 6-7 figures on average in the migration. We’re happy to help you with this - reach out to me directly!

Cortex is the best place in the world to work if you’re ambitious and want the room to run as fast as you can 😎 and we have the most fun doing it

After raising four rounds (seed -> C) I can tell you with absolute certainty that the below is true. Every extra week with the data room your chances of a term sheet drop 90%