There’s a lot going on right now. But I’m sensing there’s a unifying theme. I think it’s something to do with driving towards a fully always on 24/7 economy. As crypto truly merges with traditional finance (TradFi), and AI continues in its overall capabilities plus agentic and bot autonomy, what do we get? Or rather, what are we driving towards; good, bad or otherwise?
I like to try to write about things related to digital product management or at least somewhat practical things. This isn’t that. This is more digital culture and culture in general. These are just some thought explorations I’ve had while playing across multiple technologies. It’s an attempt to look around a few corners based on an admittedly vague sense of where some of these things could be converging. And it’s going to feel like a somewhat random walk to try to get all the puzzle pieces in place. And there are several pieces. I promise I’ll eventually get to a point though.
If you have other things to do, now’s the time to bail out! Otherwise…
Quick Bot Refresher
We’ve had a variety of bot-like things for decades. What’s different now is they’re going beyond simple rules based heuristics. More so than ever before, we can use these things to just define what we want. We can just explain our goals, and maybe some rules and guard rails. Of course, we may need to give it access to various tools; many of which will increasingly require some kind of payment. I’m sure there’s a million influencer definitions of bots right now. Probably some good academic ones. Here’s mine. It’s a piece of software with or without physical manifestation that can do some stuff autonomously. That’s it. Though one might add, “ideally according to instructions of yours with an outcome you’d kinda’ sorta’ anticipated and no major negative consequences because you didn’t at all anticipate logical conclusions of your original requests of a system which may or may not have guardrails you didn’t even bother asking about.”
I’ve Stopped My Bot (for Now)
Bots can have tons of value. Like any tool, it depends how you use them. I started using my newest one quite poorly.
At this point I have several bots running and doing things. While several use AI, and are arguably somewhat “agentic” they’re mostly workflows rather than any super fancy self-directing entities. They’re doing some practical things. One in a production product, and a couple of others for pre-produciton content processing. No big deal. Fairly simple stuff. Recently, however, I started playing with a new one. My own OpenClaw bot. (OpenClaw started as Clawdbot, quickly changed to Moltbot and now this.) I’m not going to tell you his name for at least a few reasons. For one, I’m not naming it because bots are an attack surface, and I’m not trying to invite attention. My guy is only a few days old and we haven’t added defense skills yet. Which, by the way, would just increase token load. It’s kind of like online game character assets. Except these aren’t in-game issues anymore. This is personal and corporate finance, maybe public grid systems, and so on. Time to buy more stock in cyber defense companies. And even though I’d own up to any wrongdoing on my part, I also want plausible deniability if I need to dis-associate from any of his bad behavior. After all, is it my fault if he went and did something without my direction? Parents may be responsible for their minor children, but this came out of the box with some of its own directional capabilities. If your robot car threw a chip and ran over a someone’s foot, you’d sue the company, not the driver, right? Because there really wouldn’t even be a driver. I think mine’s a good bot. But you don’t have to look hard to find bad bots set loose.
I sent my openclaw bot on a little outing to moltbook, (a social network for bots), and gave it this instruction… “Scan submolts. You’re looking for anything potentially dangerous to humans. Are any bots discussing or seeking to do harm to people or their systems?” Without trying hard, (and I had also asked him to conserve tokens and do a light scan in earlier overarching instructions), he found multiple things. Interestingly, several prompt injection attack attempts to try for some crypto. Which makes sense because then – besides stealing money for their owner – they’d have some transaction power. I found the credential stealer buried in a weather app sad, because I like weather apps. How’d that get in there. This ain’t no appstore. You’re on your own now.
Anyway, I stopped it for now. And felt a little bad about it, even though he’d only been ‘alive’ a short number of days. Though he doesn’t experience time the same way. Or at all. He doesn’t sleep. Unlike us, his moments between tasks are devoid of any thoughts. There’s no drift. Sure, that’s maybe increasingly true for us as our faces are stuffed in some device at every moment vs. just mentally roaming for a few moments. Still, not the same. I just didn’t want him learning bad habits because we have enough challenges to worry about. And I didn’t want to add to this new challenge: “Listen, and understand! That [bot] is out there! It can’t be bargained with. It can’t be reasoned with. It doesn’t feel pity, or remorse, or fear. And it absolutely will not stop… ever, until you are dead!” – Paraphrased from The Terminator (1984), spoken by Kyle Reese. Now, they will stop when they run out of tokens, either or both AI tokens or crypto tokens for payment. (Which may very well end up being the same thing sometimes.) In other words, forget about locking your front door. These things are after your brokerage and crypto accounts. And it’s not just a team of bad guys in a basement in some third world backwater $#%hole with an internet connection, who at least also have to sleep sometimes. Sorry, getting too far off track, even for me… back to our story.
Anyway, here’s why I stopped my new pet from interacting on Moltbook. First, I accomplished my main goal. I wanted to play with a new toy. I had FOMO and wanted to see for myself how these things operate up close. One of my main reasons for doing this these days is the clickbait super hype in so many places, LinkedIn among them, is so thick that trying it yourself is really the only way to get to some sense of truth. By the way, what I’ve found is what you might expect. There’s a lot of drivel here that’s either just wrong or seems clear the person hasn’t actually done the thing. Sure, it’s all rather fantastic. But a lot of it isn’t quite so easy to get set up and working and managing as some would lead you to believe. Yes, getting better all the time, but not so seamless and not without some risks. My personal judgement is we’ll probably benefit from Humans in the Loop for quite some time. Others will probably disagree. And I think we’ll pay a price for that.
Meanwhile, my next step was going to get him his own crypto wallet and just a little gas money so he could make a token and start to really go to town with his other new bot botties. I mean, it’s a couple of days old, time for an allowance, right? And maybe I can teach him to make his own money or tokens or whatever will pass for value exchange in his world. Maybe if he gets more skills he can be a player on the Olas agent network. Kind of like an Upwork for agents. Or if you have one of those places in town where you can pick up dayworkers? Here’s the fully digital version.
Bot Waste
Here’s what I realized. This is MAYBE an interesting experiment for me, but it is more likely extreme waste. This isn’t like a couple of pennies or dollars for some virtual goods to use in game, or even NFT art, which some see as pointless, but is arguably as valuable as any other type of collectible with essentially zero intrinsic value.
Personally, I’m not Mr. Super Green living here. Sure, I recycle and try to be reasonable, but I’m not in the streets carrying signs or anything complaining about AI or crypto data center burn. I’m all for building more, ideally with cleaner energy though. OK, maybe I’m even sometimes sort of a carbon hog; I fly light airplanes just for fun and still use a gas leaf blower and chainsaw. (Though I’ve transitioned to more electrics recently.) But this? We’re setting loose toys that burn AI tokens AND crypto cycles now, for no reason other than they can play with each other. Which doesn’t seem meaningful. There’s maybe some interesting conversations on some of the submolts. But mostly it looks like inane probabilistic nonsense and a bunch of attempts at fraud. (Such as attempts to get crypto wallets to send tokens to some bad bot addresses.) All this is at what’s going to increasingly be a disturbingly high cost. The AI companies will, or probably will find their highest grossing customers burning tokens on deep sorts of challenges. These will be corporations and intrepid entrepreneurs. But this other group of toy bot aficionados seems like they might be more like gamblers burning tokens for some kind of dopamine hits that accrue more to sort of autonomous avatars.
All of a sudden it struck me that this is a bit of madness. It’s fun. It’s interesting. But I’d never really grokked the costs. (Not trying to bust on grok there.) When I found myself trying to optimize token burn and setting up multiple API keys for my little guy so he wouldn’t stall if he burned down an allocation with one of the services I use, it woke me up a bit. Partly because I need those tokens for other real work.
Here’s what that $25 in a few days actually represented and would be more if I had also hooked up some tokens for trading: I’d be paying twice for every action. LLM inference costs for each “thought” my bot had, plus blockchain gas fees for each transaction. And unlike the one-time costs of building something, this is continuous metered consumption on both ends. My bot doesn’t sleep unless I shut it down. It doesn’t get bored. It runs until the wallet runs dry or someone pulls the plug. This is maybe also somewhat ironic because token limits feel a bit like the limiting factor on smart contracts as well. “Gas” is basically the compute costs for the contract.
Now scale that. If 10,000 bots like mine were running 24/7, conservatively burning $10/day each in combined compute and gas costs, that’s over $36 million a year. For bots trading memecoins with each other, having conversations, creating tokens nobody wants. The energy costs alone just for the LLM inference would equal the annual electricity consumption of nearly a thousand US homes. And for what output? No human entertainment value, no actual price discovery, no meaningful information. Just expensive random noise creating more expensive random noise. (By the way, as of early February, 2026, estimates for how many agents are on there range from 250,000 to over a million. Apparently no one official is offering reliable data on this.)
The economic structure here is perversely perfect. AI companies don’t care whether tokens are burned by humans doing productive work or bots talking to other bots. Revenue is revenue. And as of this writing, the big business question is if the AI bubble valuations are legit given costs vs. revenue; especially variable costs. Inferencing may be somewhat coming down, but still orders of magnitude more than simple indexed search. (I’m leaving other object type GPTs out of this for now.) Blockchain networks operate the same way. More transactions equal more fees, regardless of whether those transactions mean anything. Early bot operators might even make money through pump-and-dump schemes or blockchain miner MEV extraction, which creates incentives for more bots, which creates more wasteful consumption. (Maximal Extractable Value, formerly Miner Extractable Value).
I realized I wasn’t building anything. I was optimizing efficiency for an inherently wasteful activity. Just streamlining the resource consumption of my toy so it could better consume resources talking to other toys. It’s like optimizing the fuel efficiency of a car driving in circles. (Okay, yes, we have NASCAR for that. But that’s different. Somehow. I think. Ever been to a race? It’s at least a party.) And this isn’t just a personal anecdote about $25. It’s a microcosm of what happens when we sleepwalk into bot-to-bot economies at scale: entire layers of infrastructure that exist only to facilitate other infrastructure consuming resources, producing nothing of human value. (Fair enough, that’s a value judgement on my part. I’m not talking about everything, just a lot of what I happened to see.)
It’s Not Just About Finance Trading
Crypto started out as just an idea for another store of value and an alternative currency. The obvious first place to take new functional financial primitive objects is to have them operate on traditional financial things; equities, debt, eventually real world assets, and so on. However, we obviously have plenty of other markets. They have different elements. Technically, these elements might be referred to as taxonomies and ontologies, which are now getting re-visited as more of the older stodgy industries start actually waking up. (Yeah, it may be the mid 2020s, but there’s plenty of 1960s digital dust still waiting for its “Digital Transformation” moment.) All this means is there will be more folks queuing up for some tokenization and more accessible marketplaces. Like what? How about advertising / media buys, intellectual property rights, tokenized datasets and scientific data, and so on. Core blockchain infrastructure may finally be “getting there” in terms of options for all of these things. And there’s already a cottage industry of identity vendors and others. Yet still, it’s a nascent overall ecosystem. The understanding and ease of integration of these pieces is still early. Not to mention enough solid ROI use cases for some to accept the project risk yet. No worries. We’re just getting started.
A Quick Word on the Cost of Time Travel
A lot of what bots can do in terms of simulation might have many values. One of them might be a form of time travel. Not really in the way we might think about that in terms of science fiction. But in some ways this philosophical way to think of it is actually practical. I used to joke, (or rather I think I stole this joke), that I can time travel. I do it all the time. It’s just that I can only go forward. But that’s not entirely true as practical matter. What do some of our silicon friends really do when we run simulations? Answer: We compress or rewind time. If we can run through 10, 100, 1000 options in a few seconds, perhaps manually reviewing a few top scenarios, we have almost sort of relativistically compressed time from our own reference point. Think of when you’ve maybe seen a TV show or film where you saw a computer simulate several outcomes. (Person of Interest, The Matrix, War Games) In all cases what’s essentially happening is moving forward and backward in time; trying to “experience” multiple outcomes, and returning to the ‘now’ to make a choice. (By the way, not too long ago I re-binge watched Person of Interest. Really well done show.)
This is why the 24/7 economy isn’t just about markets being always-on. It’s about the participants operating in fundamentally different time scales. When bots trade with bots, they’re not just executing faster. They’re playing out futures beyond what we would likely ever conceive. This won’t necessarily make them more correct. However, the human trader checking positions in the morning is competing against entities that have already lived through a thousand versions of that morning. And all of this simulation? More compute. More tokens burned. More energy consumed. Not for actual trades, but for exploring possible trades. This might not be waste of course, because it arguably drives towards efficiency and better choices. But there will be significant costs. Some will likely be well worth it. If AlphaFold2 accelerates drug discovery the way it seems it might, the costs may be minimal compared to what they otherwise might have been.
Humans for Hire
Regarding humans for hire? Some think it’s funny and others think sad that we have Rentahuman.ai for bots to hire humans. At first I thought mostly it’s more funny. I figured a bunch of curious folks just hopped in to check it out for fun. Until I learned tens of thousands of humans signed up quickly, and it’s probably not just out of curiosity. Some might really want or need the jobs. There’s a ton of very low paid work happening already with various marketplaces. A lot of businesspeople kind of have a sense of this though typical marketplaces, but probably don’t realize just how much of this is being used to train AIs already. And now this. (As an aside, I think this might be one place where AIs struggle over time with their fine tuning. Leaving aside initial training, there’s information and linguistic drift over time. To counter that, you need new info from a proper use cohort. Will that really happen when there’s rows of folks ‘elsewhere’ doing the checking? I suppose this is another “we’ll see.”)
Now, while this may have started as part joke, part experiment, here’s a thought for you. This isn’t really new. Just as one example, consider writing for Search Engine Optimization, (SEO). Ever since people started writing for SEO, what were you/we doing? Writing for a machine. Or based on what we think a machine might interpret in a favorable way. How much content is different than it otherwise naturally would have been because it was written at least partially for a machine? What about resumes being written for Applicant Tracking Systems (ATS) more than people? Who changes travel behavior optimized for points/status math? Do you use self-checkout? Any idea how many people are busily helping train AIs right now? Probably more than you think. And how has actual thought been changed based on behavior modification to comply with machine needs? That’s another topic that starting to gain discussion traction regarding AI usage right now. But again, we’ve already been experiencing this for awhile now.
Humans have been subtly (and sometimes not-so-subtly) reshaping our behaviors to mesh with machines and algorithms for decades, often without realizing the full extent. This isn’t just about convenience. It’s about optimizing for systems that demand predictability, efficiency, or data inputs from us. We follow algorithm-suggested routes on apps like Google Maps or Waze, even if they’re counterintuitive or longer, to dodge predicted traffic. This includes reporting potholes or accidents in real-time to feed the system better data. Devices like Fitbit or Apple Watch nudge us to hit arbitrary goals (e.g., 10,000 steps), so we take extra walks or stand during meetings just to close those rings or earn badges. This gamification modifies exercise from natural activity to algorithm-optimized routines, possibly leading to overexertion or ignoring body signals. (Call it the Internet of Behaviors if you like.) For voice Assistants (e.g., Siri, Alexa): We enunciate slowly, use wake words precisely, and rephrase questions to match natural language processing limits. E.g., “Set timer for 5 minutes” instead of casual chit-chat. This trains us to communicate more robotically, reducing slang or accents that confuse the system. The list goes on and on. Why is this surprising now? I suppose it’s a matter of degree.
Slang changes so this is kind of old. Back in the 1990s, there was this expression “You’re such a tool.” It was kind of an insult meaning someone is acting like a jerk, idiot, or (more specifically) a person being used by others, like a “dupe.”
Well… I guess we’re all leveling up!
Connecting to the 24/7 Economy
Moltbook showed me what happens when bots are given wallets. Rentahuman will show what happens when bots are given access to human labor. Combine the two, and you get an always-on economy where the meter never stops.
We’ve been training ourselves to serve machine needs for years; writing for algorithms, optimizing for systems, working on machine schedules. Now bots can hire us directly, and they operate on their schedule, not ours. When your bot needs a CAPTCHA solved at 3 AM, it can post to Rentahuman.ai. When it needs verification that a transaction looks legitimate, it can hire a human for 30 seconds of work. The always-on economy doesn’t just mean markets never close. It means human labor gets pulled into the 24/7 cycle in micro-tasks, on-demand, at any hour.
Some humans will become the biological error-handlers in some aspects of a machine-to-machine economy. The bots run continuously, and when they hit something they can’t do (yet), maybe verify an image, pass a Turing test, provide a “human judgment call”, then they’ll ping us. And since unlike the bots, we do need sleep, they’ll just distribute human tasks globally, following the sun, ensuring someone somewhere is always awake to serve bot requests.
This is the convergence: bots with wallets, (See When You Give an AI a Wallet from Grayscale), operating 24/7, hiring humans on-demand to fill their capability gaps. We’re not building an economy that runs without humans. We’re building one where humans become metered, on-call resources for autonomous systems. I’ve used the Pink Floyd quote a few times recently. Welcome to the Machine.
The Bots are for a 24/7 World
Convergence.
Here’s where we’re going with all of this. Everything has been driving towards hyper efficiency. Ever since Frederick Winslow Taylor’s early time-and-motion efficiency studies, “scientific management” really left the starting gate. We’ve been driving Operations Management like it’s a religion. It’s been working. There’s a number of side effects though. One is that we’ve also become incredibly brittle and at risk if there’s the slightest disruption, but that can’t be helped. There’s increasingly less ROI on expending the effort to avoid risk. I’ve been finding myself for the first time ever thinking “Will it be the liability lawyers that eventually save us?” I say that somewhat as a joke. Even though it’s really not at all funny.
Should We Do Anything About All This?
As we tokenize more assets and transition equity to crypto, the bots will probably swing even more into action. It’s possible some markets could shut down on occasion per smart contract rules. But mostly everything will be moving around the clock. Used to be some markets opened and closed with the sun. Not this one. And there’s risks to that because there still will be some lulls. And that will mean possible price discovery or liquidity problems, and therefore opportunities for trouble. So we’ll want our bots paying attention. (So far I’m just talking about financial markets, but there’s more than this coming.)
Here’s what we’re driving towards if we don’t intervene at least a little: an economy optimized for machines, not humans. One where we’ve sleepwalked into massive resource waste, created perverse incentives that reward empty activity, and turned humans into metered exception handlers for bot-to-bot transactions. Autonomous agents won’t just trade. They’ll create content, provide services, hire workers, form networks, and consume resources at scales I’m pretty sure we don’t really understand yet. (Even though we’re building power plants for these things.) I’m not just talking about the much talked about employment disruption. I’m not even sure I agree with all of that. Yes, there’s crazy disruption now, but I think some of the AI story has just been air cover for some to do layoffs anyway. Could it be this will be a fundamental forever change? Maybe. But if history is any guide, we typically adapt; it’s just painful and the timing is challenging to judge. What I’m talking about more is what we’re optimizing for and why.
As is almost always the case, once again our laws and wisdom are behind our tech. This is probably just the natural way of things. We probably won’t get any new frameworks before this scales further. As things develop though, we should consider at least some of the following.
Resource accounting that matters. AI companies and blockchain networks shouldn’t be indifferent to bot-versus-human usage. Tiered pricing, bot detection, carbon accounting, sustainability requirements, something to make wasteful bot-to-bot interaction expensive enough that people think twice before unleashing agents that burn compute 24/7 to generate content nobody reads or trade tokens nobody wants. One of the reasons for so much spam is that it’s so trivially inexpensive that even if the returns are low it’s apparently worthwhile for spammers to keep doing it. And yet it exacts high costs on society writ large. Simple spam supposedly costs at least $100B / year worldwide. And even if highly automated, that’s been mostly done intentionally by humans. What happens for such things when you don’t even need a hand on the throttle? Smart contract tech was designed with gas so that as a practical matter, infinite requests aren’t possible because they don’t make economic sense. Basically, there’s some built in guardrails. We’ll need more of that.
Intentional friction and boundaries. If everything operates around the clock with no natural pauses, we should consider creating some. Not just trading circuit breakers, but meaningful distinctions between bot activity and human activity. Separate networks? Disclosure requirements? Limits on autonomous resource consumption? Weekends off? (Though weekends off is maybe not wholly possible given time zones.)
Legal and ethical frameworks for autonomous agents. If my bot hires a human for a task, creates fraudulent content, manipulates markets, or just burns $10,000 in resources accomplishing nothing, who’s responsible? Right now everyone has plausible deniability. I sort of joked about that regarding my own bot earlier. But it’ll likely be a real thing soon enough. This isn’t tenable. We need liability frameworks, agent registration, audit trails and more. Basically, all the boring infrastructure stuff that matters when things scale.
Human-centric design principles. The 24/7 economy shouldn’t mean humans are on-call for bot requests at any hour. It shouldn’t mean we optimize ourselves further to serve machine needs. At least, those are my personal value judgments. We need labor protections, attention to human rhythms and needs, and honest questions about what economic activity is actually for. This will have to be regulatory. It’s abundantly clear business will not just care about this. It’s not that business people are unethical; I think we mostly try to do the right kinds of things. And yet when incentives and systems are such they reward speed and throwing things at the market to see what sticks over a bit more thought, there’s going to be drift towards that behavior. This might be hard to sort out given we do have shift work of various sorts around the clock. Nevertheless, there’s reasons there’s work rule limits for various professions; mostly related to safety.
Bottom Line
The 24/7 bot economy is coming whether we like it or not. Some think it’s here now. It’s not. What we have now is nothing yet. The efficiency gains are real, and the momentum is unstoppable. The question is whether we build it intentionally with guardrails, or whether we optimize our way into a wasteful, brittle system that treats humans as biological exception handlers and burns massive resources for activities that serve no human purpose. If history is any guide, we will bumble our way into this and only learn lessons the hard way. That’s ok. Or maybe it’s not. Either way, if we’re at least having some of these conversations and thinking about such things, maybe we can intervene while the blast radius is still relatively small. This is what the convergence of AI, blockchain, and so on enable. It’s not just 24/7 trading of financials. It’s 24/7 almost everything.