The future of the world’s economy and defenses will be based on SWARMS Those who can organize and direct swarms of agents controlling intelligence and swarms of physical machines will be unlike any power we’ve seen in history

I’ve tried most agent platforms, and Hermes remains the most powerful The market for agents is huge and there will be a range of options But for those who want the most capability, customizability, and sovereignty, Hermes is the one

We’re in the AI agent “honeymoon period” Many are experiencing this using Grok Bot, Instinct, Muse, or Dots for the first time It feels amazing at first - “holy sh*t it just did this thing for me!” Everyone races to tell their friends, it’s the talk of social gatherings - “dude my Instinct got me a refund and all I did was text it” You start doing more and more with it, handing it over more data so it can be even better at helping you But then something happens - it misses an important detail, it screws up something you have to fix, it just feels not as good anymore Anyone who’s been using agents since the early OpenClaw and Hermes days knows this feeling The only question is what do you do? Do you “break up” with your agent and try another one, or do you keep going with the faith that it will get better with time (it will), or do you stop using agents and submit to the permanent underclass?

Just to show how early we are with enterprise usage of AI Within enterprises using OpenAI, the vast majority of users are not even using plugins or skills yet Even at the top 10% of companies by AI usage, only about 1 in 5 weekly active users touches plugins or skills in a given week So to conclude: 1. Inference spend at enterprises is still just a rounding error in aggregate (0.1%) 2. Most enterprise users of AI are not even using plugins or skills yet We have so much room the grow with AI in the enterprise

This chart by @a16z is the macro side of the thesis at @codos_ai Despite all the talk about the cost of tokens, inference spend is still just a rounding error in aggregate But we believe this continues to rise significantly in the coming years, and enterprises will need transformation partners to guide the ensuing change that comes as usage and inference spend grows to larger levels of total enterprise spend

Here's an example of how we're using Hermes agents inside @codos_ai connected to our company brain The brain has all the context, so at any given time I can just ask it something like "who are the top 3 FDE candidates that I should reach out to?" I can ask via Hermes, it pings the company brain via MCP which has all the info, pulls back the results with next action, then I can have Hermes draft the emails to schedule the interviews Codos brain + Hermes arms and legs!

It’s pretty remarkable to see just how early we still are And if we’re this compute constrained already, what happens when we go from 2% to 10% of households, which isn’t unreasonable over the next 12-18 months

It’s been wild to see how much inbound interest we’ve received after the public launch of @codos_ai But also surprising to see some of the behavior from the other side of the table from where I’ve been last 5 years @dimakhanarin taking the brunt of it

I'm a big believer in owned intelligence, especially for enterprises, and if you're running agents - you need to own the trace data Why? Because that data can be used to train models that you own based on your data I've been waiting to check out @OvermindLab because their platform makes it easy to turn an agent’s real production behavior into specialized models that you owns

Quick tip - almost every SaaS service you use will automatically offer you a lower price if you go through the cancellation workflow Just before the final step is usually when the best offer appears, have your agents do this for you and save money I did fully cancel this one

AI in small business is easy - you can just do things and should be using it everywhere possible in your company Once you get to a certain size, 50-100 people, it starts to get more complicated Legacy systems, approval chains, company behavior norms - it starts to require more change mgmt, harder to just do things but can be done with a strong CEO/owner mandate The hardest spot is actually mid-enterprise, companies in the 300-3,000 people range These are organizations that typically don’t have dedicated AI teams or a Chief AI Officer like most large enterprises and don't have the size and/or resources to hire the big consultancies and deploycos to run AI transformation programs They're big enough to realize sizable impact from AI, but can't afford to staff up teams to run pilots and dedicate the management cycles to develop thoughtful AI strategies - they're too busy trying to run the business This is the market segment that I'm most bullish on serving at @codos_ai We come in, run a short diagnostic that includes both business strategy and AI, outline a prioritized list of high impact initiatives, then run the deployments alongside the existing team What these companies get is a "virtual" Chief AI Officer underpinned by a benchmark-winning company brain context layer that enables AI automation across the company And we deliver it at a fraction of the cost of what a mid-enterprise would pay to a large consultancy or deployco

One of the great things about working at a startup is we get to do everything in an AI-native way No legacy tools, no cumbersome processes, just design it the way it should work in the age of AI We built our hiring process this way: >applicants apply on website >data is ingested into the ATS we built on top of our company brain (no need for external software) >agent reviews each applicant, scores against a rubric, updates the candidate card, decisions those that don't need human judgement >agent pings one of us in Slack for human actions needed >each morning agent executes a workflow that sends a personalized daily report with prep for interviews on the calendar and any outstanding actions Then at any time the Codos company brain can be queried for status on any applicant, metrics on any job posting, or anything related to hiring and all the dots will be connected This is the way

Something I wonder about: What % of remote employees at large companies are using AI to free up time in their day, then using that free time to do things around the house vs do more things for the company?

I am so incredibly convinced: 1. AI is existential to most businesses >not adopting AI is the worst game-theoretical option - see my pinned post if you want a full explanation 2. Most companies are barely scratching the surface today >despite what you read on this app, the most common implementation today is a frontier enterprise account where employees use it for better search and some basic automations, mostly in single-player mode 3. AI is moving so fast that it’s overwhelming for most enterprises >most are not sitting on X all day and on top of the latest model releases and tool announcements, they have actual day jobs and can’t keep up or don’t want to, they want someone to tell them what to do and just make their PnL and KPIs better 4. Most mid-enterprises and a good chunk of large enterprises still don’t have a real “head of AI” or an enterprise AI strategy >they have internal champions and fragmented deployments, not an interconnected strategy and plan for how AI should be used across the business 5. Most enterprises have not connected AI to the PnL >giving AI tools to employees can definitely improve productivity, but what do they do with that extra time that benefits the company? That’s the link that’s often missing 6. There has never been a better market opportunity for service providers than this AI transformation wave >all previous “enterprise change” catalysts - dotcom, mobile, cloud - pale in comparison to the upending nature of what AI will do because AI requires a fundamental rethink of how you organize inputs to produce outputs At the very least enterprises will have to redesign processes with AI at the core, and on the other end of the spectrum many enterprises will need to upend their entire operating and business models or risk becoming irrelevant This is excruciatingly hard work because you have to keep the business running while all this redesign and rethink is going on, and that’s why the right partners who bring business strategy + frontier AI + execution will crush

Wait, what? “Compute demand is doubling roughly every week” (for Instinct) And rumors that Meta is scrambling to add compute for Muse as growth is off the charts Once again, I implore you to long your compute longs

Update: @runwayml is a complete rip off for video generation $200 in credits later and it’s output is unusable Someone convince me otherwise Maybe it’s a skill issue but I’m done paying their insane tuition

It's been over a week since the Jev launch and I'm convinced this wasn't one of those "new shiny objects" that quickly fades into irrelevance like many hyped launches Jev is legit a key component of the playbook for efficient model use, believe its here to stay