When we look back, Alex Karp may have initiated an important preference cascade around AI sovereignty. It’s worth noting what @Benioff and @satyanadella are both saying: Your knowledge, as a company, is your sovereignty. If you lose it to someone else (anyone else) you are hollowing your organization out. There are many ways to accidentally leak intelligence so you need partners and tools who can sign up for the complexity required to give it to you. See Benioff below and see Satya’s essay linked below.

Own Your Intelligence.

TL;DR: Use a control plane, pick your model on a per task basis, do it massively cheaper as a result, keep your edge, don’t leak it to the frontier labs. The thoughts below are exactly what we find in large enterprises that use 8090’s Software Factory control plane. Our control plane is agnostic and sits above all the model chaos. You use it to manage your engineering team. It creates a more rigid conformation to the software development lifecycle so Owners and executives get well documented, governed, auditable software - not broken promises, vaporware pr large T&M bills. Engineers need to do a bit more work upfront but then are relieved of dealing with infinite slop in return. For example, on a typical migration task of an enterprise workload (PHP—>next.js), when we chose Opus 4.8 inside our control plane, Software Factory gets to the right answer 1.5x faster and 1.4x cheaper than using Opus alone. If an Enterprise asks our control plane to use an open weight model like GLM, instead, the costs fall even more dramatically. The migration is 16.4x cheaper but 3x slower but still works! This brings up the obvious: at some near point in the future all large enterprises - ie those currently spending millions per month or more on tokens so they seem brilliant to a Wall Street analyst or their Board - will have to rationalize why the would spend so much more for inefficiency, poorer outcomes and data leakage to providers that increasingly also want to compete with them.

Meta burns $2.65B a year on AI tokens. at $300K for a Meta engineer, that's enough to pay ~9,000 engineers for a full year. now ask yourself: since the layoffs, has Meta shipped anything that feels like 9,000 engineers’ worth of output?

We raised a $135M Series A! 8090’s Series A was led by Salesforce Ventures and joined by WNDR, Craft Ventures, The Production Board, and LAUNCH. We also had the support of a group of esteemed angels including Nikesh Arora, Cliff Robbins, Adam D’Angelo, Shyam Ravindran, Abhi Arun, and Thomas Laffont. We’re grateful for their support. It validates 8090’s mission and traction so far, but mostly it accelerates the work ahead. The capital will go to two places. The first is hiring more people, because the demand we have is accelerating rapidly. The second is investing in the compute and infrastructure needed to keep delivering our solutions at high quality and reliability. 8090 works with the biggest, hardest, most demanding customers in the most regulated industries: healthcare, insurance, life sciences, aerospace, energy, manufacturing, financial services, and the United States government. We help them win by using our AI-enabled Software Factory to design and build entire new systems, refactor old ones, and find and accelerate their edge. Our view is that as Software Factory is used more and more to do mission-critical work inside industries with the least tolerance for error and the most oversight, it will be used to bring transparency, consistency and control to work everywhere. And as we expand the potential of the biggest organizations, we are also building a playbook and a series of network effects into Software Factory that will be valuable to everyone, from SMBs to solo founders. With much gratitude, back to work… PS - A note on why I am doing this as CEO, rather than from the board. This is one of those rare moments when the technological ground is moving so ferociously underneath all of us that the decisions made in the next few years will set the stage for the next twenty. AI can be the grand equalizer. It is the thing that can give everybody a shot, and I would like to help it achieve that potential. Since I left Facebook, I was waiting for a moment like this to return to a full-time operating role. I was a demanding manager back then, but I felt I had no choice given how powerful and undeniable what we were building was. I am convinced that what we are building now is even more important, so there was no decision to make except to be all in.

At @coinbase our AI spend is down nearly half this quarter while token usage keeps climbing. My team built the infrastructure behind it: routing, caching, cheaper defaults, and the spend services that track it. We route everything through our own gateway: a single endpoint and format for dozens of models, with cross-provider failover, redaction, logging, and cost controls all applied before anything reaches a vendor. We started with cheaper defaults and caching. 91% of employees weren't hitting their usage caps. Instead of lowering caps, we set cheaper model defaults to cut spend. Caching took more work to get consistent across every tool and model family. A cache hit needs the prefix to match exactly, so we keep building a long, stable prefix across turns. Each request only pays full rate on the new tokens and reads the rest from cache. Our routing accounts for caching too. The naive approach scores each turn on its own and sends it to whichever model fits, which seems reasonable but would run up spend. The cache is per-model, so switching mid-conversation invalidates it. Our router weighs cache state alongside how hard the task is: a conversation keeps its model while the cache is warm, and the chance to re-route comes only when it goes quiet long enough for the TTL to lapse. Once it does, the router is free again to pick the best model for the task. These improvements happened at the gateway, so they apply across every team and tool. Next we're going deeper on the coding harness, where we have the most signal and flexibility, tuning how subagents and context get managed.

How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching. Better Defaults (not Usage Caps) – Engineers can choose any model they want, but defaults matter. We’re experimenting with defaulting to open weight models like GLM 5.2 and Kimi 2.7 through our LLM gateway, while still encouraging engineers to choose the right model for the task. 91% of our employees were never hitting their usage caps, so instead of lowering caps and driving up alerts, we're moving to cheaper defaults. Note that code reviews use a diversity of models, so they can check each other's work. Better Routing – In our custom harnesses, we preprocess prompts and route to the best model for the job, considering cache hits and model pricing. For instance, you may want a frontier model for planning, but not for execution where they can be overkill. Ultimately, humans shouldn't be choosing models - AI can automate this task. Better Caching – Cache misses are the easiest way to drive your cost up. All of our requests are cache aware, so we’re reusing a warm cache wherever possible. For example, our cache hit rate went from 5% → 60% in LibreChat once properly implemented. Keep Context Lean – Start fresh sessions when switching tasks. Scope file context narrowly. Disconnect unused tools. Don't just compact. The goal isn't fewer tokens used, it's fewer tokens wasted. Better Visibility – Our engineers can use as many tokens as they want, from whatever model they want, but we’ve made usage visible – and the more you spend on AI, the more impact we expect. The goal isn't to suppress usage. It's to build the infrastructure that makes exponential growth sustainable. Putting this into practice has cut our AI spend nearly in half, while our token usage continues to grow.