Always-on agents just hit the limits of compute, cost, and control
by Tony Thomas
The OpenClaw situation is a warning shot for enthusiasts of always-on agents.
On April 4, 2026, Anthropic changed how Claude subscriptions work with third-party harnesses such as OpenClaw. They no longer cover unlimited use of agents on flat-rate plans. Sure, you can still run them, but now you must pay through API billing or usage bundles. Anthropic said the increased use of OpenClaw and similar harnesses placed strain on their systems and did not resemble normal subscription traffic.
Why OpenClaw Became So Popular
OpenClaw and its derivatives took off because they turn models into agentic workers. They run locally, connect to frontier models, and execute tasks across apps and time zones. The emphasis is on persistence rather than chat or one-off prompts. Even when you are sleeping, your squad of agents keeps working in the background. That autonomy makes it expensive.
While a chat stops, an agent keeps going. It plans, retries, summarizes, calls tools, and loops. Left alone, it keeps burning tokens at a steady rate. Flat subscriptions were built for short bursts, not waves of activity, so the mismatch is inevitable.
The OpenAI Hire Changes the Context
In February 2026, OpenClaw’s creator, Peter Steinberger, joined OpenAI to work on personal agents. Around the same time, OpenClaw moved to an independent open source foundation supported by OpenAI. Weeks later, Anthropic removed OpenClaw and similar harnesses from subscription coverage. The timing adds context, and the sequence is hard to ignore.
Third-party harnesses sit comfortably between users and models. They decide when to call the model and how long to loop. That makes demand unpredictable. Metered pricing restores control. It also shifts profit back to the labs.
Subscription Pricing vs Agent Reality
Of course, this extends beyond OpenClaw. Other third-party harnesses still work, but they no longer ride on flat plans, and usage is metered. Continuous agent use now requires continuous payments.
Agent workflows scale with time. Leave them running, and they keep consuming compute. Tools that manage email, calendars, and tasks generate a steady load. Infrastructure built for interactive requests handles spikes. It struggles with constant demand. And there is only so much compute to go around.
Flat pricing assumes pauses. Agents remove pauses. That is the core conflict.
The Reaction From the Community
Part of what pushed me to write this was the reaction from developers building agent-driven workflows. A video from Network Chuck captured the frustration. He described receiving notice that subscription coverage for tools like OpenClaw was ending. Many users chose Claude for that flexibility.
Chuck also suggested the move came down to infrastructure strain or subsidized usage. His tone was jarring. These are no longer experiments. People are wiring agents into daily systems. Pricing changes are a gut punch. And it puts strain on people’s businesses and their lives.
Platform Control Is the Bigger Story
This also looks like a shift toward increased platform restraint. Labs are not just shipping models anymore. They are building full agent stacks that include orchestration, tools, scheduling, and memory. The trend is now in creating packaged solutions rather than just doling out compute.
Third-party harnesses sit outside that stack. They drive loops that the provider can no longer control. Moving orchestration inside the platform limits runaway usage and stabilizes demand. It also centralizes the agent layer.
Security and Autonomy Tradeoffs
Autonomous agents often need wide system access. They read files, trigger actions, and move across apps. That expands risk. Prompt injection becomes more dangerous. Automation mistakes scale faster.
Data exposure increases. And runaway agents can burn tokens and suck up compute at alarming rates. Provider-controlled agents reduce some of that surface area. Less freedom. More guardrails.
The Local Hardware Escape Hatch Is Closing
Local inference once looked like the perfect workaround. You could buy hardware, run open models, and skip token costs.
That window narrowed quickly. GPU demand surged. High-VRAM cards became expensive. Plus, larger setups require power, cooling, and upkeep. Lots of power. Even modest rigs add ongoing cost.
As a result, local inference stops looking like a lower-cost alternative. It becomes a large capital expenditure plus ongoing operating expenses. Continuous agents amplify both.
Between Cloud Tokens and Local Hardware
This leaves an awkward middle path.
Cloud models get expensive when they run constantly, and local hardware that can replace them is also expensive. Neither side offers a cheap solution.
Agentic workflows push requirements higher. Longer context. More frequent calls. Increased tool usage. Small local models can route or filter, but full autonomy still requires frontier models.
So the compromise is a hybrid. You run lightweight steps locally, call frontier models when needed, add budgets, and limit loops. It works, but autonomy shrinks.
Using fully local agents becomes harder. Cheap cloud agents are fading. That is the dilemma.
The End of Cheap Unlimited Agents
This is not the end of agentic AI. It is the end of cheap, unlimited agents.
OpenClaw still runs. It just pays for what it uses. If it runs all day. And the meter runs all day.
That pushes designs toward boundaries such as budgets, step limits, and event-driven triggers.
Hybrid stacks become more common, with smaller models handling planning and frontier models handling heavier reasoning.
The goal shifts from autonomy to efficiency.
What Happens Next
The OpenAI hire underscores how central orchestration has become. Agentic control is now the center of the red-hot competition between labs.
OpenClaw did not fail. It just exposed the economics. Always-on agents turn models into infrastructure. Infrastructure costs money. Subscriptions blur that cost. Metering makes it visible.
Expect more tightening, higher prime time rates, fewer open loops, more bounded agents, and more hybrid setups.
Letting agents run forever in the background at a massive scale was always untenable. And the math finally caught up.
I use OpenAI’s tools every day. They sit at the center of my workflow. They help me think, draft, outline, code, and ship ideas faster than I ever could alone. I am grateful for that. I am also uneasy.
Here is the tension as I understand it. The company generates roughly $3.5 billion in annual revenue. It also burns somewhere between $5 and $7 billion a year. That gap would sink most businesses. OpenAI bridges it by raising more capital. It has already raised about $6.6 billion at a reported $157 billion valuation and is seeking another $15 to $25 billion, with SoftBank mentioned as a potential anchor. The growth is real. The cash burn is just as real.
As someone who has watched tech cycles for decades, I know this pattern. Big vision. Big spending. Big expectations. Sometimes it ends in dominance. Sometimes it ends in a write-down.
The Cost of Intelligence
Training frontier models like GPT-5 requires massive clusters of specialized chips running for months. Industry estimates place training costs in the hundreds of millions per generation. Future systems could cross the billion-dollar mark. That is serious money even in Silicon Valley.
Inference adds another layer. Every prompt I send consumes compute. Multiply that by millions of users and enterprise workloads, and the meter runs constantly. Efficiency improves, yes. But usage tends to expand faster than efficiency gains. I know this because my own usage keeps climbing.
Lower API prices complicate the picture. Competition has pushed prices down. Enterprises negotiate volume discounts. Developers design apps that can switch between models. Lower prices expand adoption. Higher adoption inflates compute bills. The flywheel spins both ways.
What This Means for the Average User
If you are an everyday user, you may not care about burn rates. You care about reliability, pricing, and whether your favorite features stick around.
Here are the practical implications. Prices could rise if capital tightens. Generous free tiers could shrink. Enterprise features may get prioritized over hobbyist use. Rate limits might tighten during peak demand. Product packaging could shift as the company experiments with margins.
On the positive side, pressure can drive innovation. We may see faster, cheaper models. More specialized tools. Better optimization that reduces latency and cost. Financial strain often forces discipline. Discipline can lead to better products.
Still, the era of “growth at any cost” rarely lasts forever. At some point, someone insists on a profit.
The Microsoft Factor
Microsoft has committed an estimated $13 billion, much of it in Azure cloud credits. OpenAI depends on Microsoft’s infrastructure to train and serve models. Microsoft integrates OpenAI into Copilot and Azure offerings.
Microsoft has options. It can subsidize AI losses with profits from enterprise software and cloud. It can build internal models. It can renegotiate commercial terms over time. OpenAI does not have the same flexibility. AI is the business.
That leads to a question many users quietly ask. Could Microsoft eventually acquire OpenAI outright?
Takeover Scenarios
A full Microsoft acquisition is plausible in a long-term squeeze. If capital markets cool and OpenAI needs stability, Microsoft would be the obvious buyer. The infrastructure is already intertwined. The product integrations are deep. A clean acquisition could simplify governance and funding.
Nvidia is another interesting possibility. Nvidia controls the chips that make modern AI possible. Owning a leading model provider would give it vertical integration from silicon to software. That said, regulators would look closely at such a move. Nvidia already sits at the center of the AI supply chain.
There is also the possibility of a broader tech consortium, or even a large sovereign-backed investment group, stepping in if AI becomes viewed as strategic infrastructure.
As a user, I try to imagine what each outcome would mean. Under Microsoft, OpenAI might become more enterprise-focused and tightly integrated into Windows and Azure. Under Nvidia, optimization for its hardware would likely intensify. Under a consortium, priorities could fragment.
Independence gives OpenAI flexibility. Acquisition could give it stability.
Will the Models Get Smaller?
Another question I find myself asking is technical. Will OpenAI continue building ever-larger frontier models? Or will it pivot toward smaller mixture-of-experts systems and highly specialized models?
Mixture-of-experts architectures allow parts of a model to activate selectively. That reduces compute per query. Smaller, specialized models can handle narrow tasks at lower cost. In a world where inference costs matter more than leaderboard scores, that strategy makes sense.
I would not be surprised to see a portfolio approach by 2027. One or two flagship frontier models for headline capability. A suite of efficient MoE models tuned for coding, legal analysis, design, or customer support. Perhaps even on-device variants for certain use cases.
If cost pressure intensifies, efficiency will not be optional. It will be survival.
Three Possible Outcomes
From where I sit, there are at least three broad outcomes.
First, OpenAI achieves technical breakthroughs that materially widen the gap. It bends the cost curve through architecture and hardware optimization. Revenue scales faster than expenses. The company grows into its valuation and becomes the central platform for AI services. In this scenario, today’s burn looks like an investment phase.
Second, competition compresses margins. Performance converges. Pricing power erodes. OpenAI remains important but becomes one strong provider among several. Growth slows. Valuations reset. The company either restructures for efficiency or accepts acquisition.
Third, a major strategic shift occurs. OpenAI leans heavily into enterprise software, builds a durable recurring revenue base, and looks more like a next-generation enterprise platform than a pure research lab. It may not dominate consumer mindshare, but it becomes deeply embedded in business infrastructure.
None of these outcomes feels impossible.
What 2030 Might Look Like
By 2030, I suspect OpenAI will look less like a research lab chasing the next giant model and more like a layered AI platform company.
There will likely be a flagship general model, yes. But around it, I expect specialized vertical models, optimized inference stacks, custom enterprise deployments, and tighter hardware integration. The business mix may tilt heavily toward enterprise contracts rather than consumer subscriptions.
Governance will probably be simpler. Investors will demand it. Whether that means a clean for-profit structure or ownership by a larger parent, I doubt the hybrid era lasts unchanged.
As a user, I hope the culture of rapid iteration survives. I have benefited from that speed. I also hope the economics mature. I would prefer a boringly profitable OpenAI in 2030 over a dazzling but fragile one.
Why I Care
I have skin in the game. My workflows, my projects, and even parts of my income rely on these tools. They have multiplied my output in ways I did not think possible a few years ago. I am a happy customer.
I am also a cautious one.
The financial paradox will not resolve through vision alone. OpenAI has to translate technical leadership into durable economics. If it bends the cost curve and builds defensible advantages, it could define the next decade of computing. If it cannot, capital markets will eventually enforce discipline.
I have seen enough cycles to know this: brilliance buys time. Sustainable cash flow buys permanence. I am rooting for both.
UPDATE: The Pentagon Deal, the Backlash, and a New Twist
Just when I thought the OpenAI story could not get more layered, the past few days added fuel to the fire.
On February 28, 2026, OpenAI finalized a $200 million contract with what is now being called the Department of War, formerly the Department of Defense. The agreement allows OpenAI’s models to operate inside classified military networks. The timing raised eyebrows because it came just hours after the Trump administration cut ties with Anthropic. Within days, social media filled with calls to “Cancel ChatGPT,” driven by concerns about AI militarization and surveillance.
As someone who uses these tools every day, I felt the tension immediately. Pride that the technology has reached that level of national importance. Unease about where that road can lead.
What the Pentagon Deal Actually Means
OpenAI says the contract includes clear limits. No autonomous weapons. No domestic mass surveillance. Those lines matter, and I take them seriously.
Critics point to language suggesting the models could be used for “all lawful purposes.” That phrase carries weight. Under existing laws, “lawful” can include broad data collection. That is where trust becomes fragile.
Sam Altman acknowledged the deal was rushed and admitted the optics were not great, especially given the immediate fallout with Anthropic. I appreciate that level of honesty. Still, once a company steps into defense infrastructure, the public conversation changes. This is no longer just about chatbots and coding assistants.
The Anthropic Fallout
Anthropic reportedly refused to grant the Pentagon unrestricted access to its Claude models, citing safety concerns. The administration responded forcefully.
President Trump ordered federal agencies to stop using Anthropic technology. Secretary of Defense Pete Hegseth labeled Anthropic a “supply-chain risk,” language that usually carries serious national security implications. In one stroke, a leading AI firm was cut out of federal work.
Here is the interesting twist. Despite the federal ban, Anthropic’s Claude shot to the top spot in the App Store shortly after the announcement. Users responded quickly. If anything, the controversy appears to have boosted public interest.
That tells me something important. The AI market is not just shaped by contracts and capital. It is shaped by public sentiment. In a consumer-driven ecosystem, backlash can move download charts overnight.
The Money Keeps Climbing
At nearly the same time as the Pentagon deal, OpenAI announced a $110 billion funding round led by Amazon, Nvidia, and Microsoft. The valuation now sits around $730 billion. That is an extraordinary number for a private company.
Financially, it signals confidence from some of the most powerful players in tech. Symbolically, it raises the stakes. Investors at that level expect sustained growth, expanding margins, and strategic leverage. This is no longer a scrappy startup narrative.
Some critics, including a few employees, argue that the company is drifting from its earlier safety-first framing toward political alignment and profit maximization. I cannot see inside those conversations. I can say that when the capital stack gets that large, the pressure changes.
What This Means for Everyday Users
If you are an average user, you might wonder how this affects you.
In the short term, the product likely continues to improve. Massive funding means more compute, more research, more engineers. Government contracts can provide stable revenue. Enterprise adoption will probably expand.
In the longer term, priorities may shift. Defense and enterprise needs could shape model capabilities. Compliance requirements could influence product design. Pricing structures could evolve as the company balances public access with large institutional contracts.
Meanwhile, the competitive field is wide open. Claude climbing to the top of the App Store shows that users are willing to move. Multi-model strategies are becoming common. Loyalty now depends on performance, trust, and cost.
My Personal Take
I remain a happy user. These tools have changed how I work. They have multiplied my output and sharpened my thinking. I would not want to go back.
At the same time, I am more aware that OpenAI now sits at the intersection of geopolitics, capital markets, and national defense. That is not a comfortable place to stand. Companies in that position face scrutiny from every direction.
By 2030, OpenAI may look less like a fast-moving lab and more like critical infrastructure. It may operate with tighter governance, deeper enterprise roots, and closer alignment with major tech partners. Or it may become part of a larger corporate structure entirely.
Brilliance got it here. Public trust, disciplined economics, and careful power management will determine what comes next. As someone with skin in the game, I am watching closely.
