The Governed Machine: SolarWinds Avior and the Trust Layer Crypto Never Shipped

Five vendors. Forty-eight hours. The same quiet sentence, repeated like a prayer in five different boardrooms. In the last week of a bull market that has made everyone forget how to be afraid, the enterprise software industry released its fifth agentic AI system — SolarWinds Avior — and the timing was not lost on me. Not because the product was remarkable. Because of what its chief technology officer said while describing it. He said the system builds probabilistic models on top of deterministic execution. The AI decides what to do. A fixed framework — predefined goals, approval gates — decides how. I read that line three times. It was, almost word for word, the promise blockchain made to the world in 2015 and then quietly broke. The code whispers, but the soul listens. And somewhere in a crowded market of autonomous agents, an old confession was being made in a language the crypto industry has forgotten how to speak. To understand why this matters to anyone holding a hardware wallet, you have to understand what Avior actually is. SolarWinds — the company whose name became shorthand for the worst supply chain breach in enterprise history — has shipped an agentic AI system for IT operations. It merges alerts, detects configuration drift, restores broken configs, cleans up inventory. Unglamorous work. The kind of work that keeps hospitals and banks running at 3 a.m. The company has built it around three pillars — goal orientation, transparency, and control — and three operating modes that scale from "assist me" to "act on my behalf" to "act alone." It supports the Model Context Protocol, the emerging standard for how AI agents talk to tools. And it is sold, not to strangers, but to the 300,000 customers SolarWinds already has. One configuration switch. No new relationship required. Here is what is missing, and it is the most important part. SolarWinds published no efficacy data. No adoption numbers. No measured reduction in mean time to resolution. No customer case study with a before and after. In the same week, a competitor named DXC published 500 use cases and a 95 percent reduction in resolution time. The industry can produce evidence. SolarWinds chose not to. And beneath the launch sits a history that no amount of governance marketing can erase: in 2020, SolarWinds itself was the vector. Russian state actors compromised its build pipeline and pushed malicious code to 18,000 customers, including the United States Treasury. The company now sells a product whose central promise is controlled, auditable, reversible autonomy. The irony is not incidental. It is the whole story. The first thing that struck me about Avior's architecture is that it is a smart contract wearing a suit. The CTO's formula — the model decides what, the framework decides how — is the exact division of labor that Ethereum encoded into the EVM a decade ago. A smart contract is a deterministic execution environment: given the same inputs, it always produces the same outputs. It cannot hallucinate. It cannot improvise. It does precisely what its bytecode says, and nothing more. The entire value proposition of the blockchain, stripped of ideology, was this: a place where the machine cannot lie because the machine cannot deviate. Avior has rediscovered that proposition. It has wrapped an unreliable, probabilistic model in a deterministic cage and called it governance. I have been auditing code for most of my adult life, and I want to be precise about what this is: it is not innovation. It is rediscovery. And rediscovery, in this industry, is often more revealing than invention, because it tells you which problems are actually real. Consider the oracle problem. When Avior's framework decides "how" to act, it does so according to predefined goals and approval gates. But who defines those goals? Who writes the policy? Who decides which operations are dangerous and which are safe? The moment you ask that question, you have left the clean world of deterministic execution and re-entered the messy world of human judgment. This is the same wall crypto hit in 2016, when The DAO was drained not because the code was wrong but because the humans who wrote the code failed to imagine an attack their own logic permitted. The code was deterministic. The humans were not. Avior's three pillars — goal orientation, transparency, control — are an admission that determinism alone is insufficient. You need a policy layer. You need a human in the loop. You need, in other words, exactly the governance scaffolding that every serious protocol has been forced to build, usually after the first exploit. The transparency pillar deserves particular scrutiny, because transparency is where crypto and enterprise AI both lie to themselves. Avior promises an auditable record of what the agent did. But auditability is not the same as explainability. A deterministic framework can log every action the agent took — its trajectory — while remaining completely opaque about why the underlying model proposed that action in the first place. This is the difference between a flight recorder and a mind reader. Crypto has spent years marketing the flight recorder as a mind reader: on-chain transparency that reveals every transaction while concealing every intention. In my 2020 retreat, when I analyzed fifty DeFi contracts during the isolation of that strange summer, I found the same pattern over and over — protocols that logged everything and explained nothing, where the transparent governance was a stage on which a handful of insiders made decisions the community could watch but never influence. Transparency, I learned, is the most honest ledger only when someone is accountable for what it records. Now consider the interoperability layer. Avior supports the Model Context Protocol — the emerging standard for how AI agents connect to tools and data sources. If you have watched crypto for a decade, this should feel like déjà vu. MCP is ERC-20 for tool calls. It is a common language that lets agents from different vendors talk to systems that were never designed to talk to each other. The standardization is genuinely useful, and it is also a new attack surface. When every agent speaks the same protocol, a single poisoned tool becomes a vector into every agent that trusts it. This is the supply chain problem in miniature, and it should terrify anyone who remembers what happened to SolarWinds in 2020. The company that is now selling governed autonomy was itself the weapon. Russian actors compromised its build pipeline and delivered malicious code to 18,000 customers through a trusted update channel. The lesson of that breach was never that SolarWinds was careless. The lesson was that trust, in software, is transitive and therefore fragile. You do not have to breach the fortress. You only have to become a trusted supplier to it. And here is where the crypto industry should feel the ground move beneath its feet. We tell ourselves that we solved the supply chain problem — that immutability, that on-chain verification, that trustless systems make a SUNBURST-style compromise impossible. We built towers of glass on beds of sand. The bridge hacks, the npm backdoors, the compromised MCP servers, the liquidity pools drained through a single malicious dependency — these are the same attack, wearing a different coat. The code was deterministic. The trust was transitive. The humans, again, were the vulnerability. Avior's answer is a safety net: automatic rollback, restoration to the last known good configuration. I read that feature and felt a quiet sadness, because a rollback mechanism is a confession. It says, in plain engineering language, that the agent will sometimes do the wrong thing, and we have built a way to undo it. Crypto has no rollback. That is its virtue and its curse. When a bridge is drained, the ledger remembers forever, and the users are told that this is what trustlessness means. Avior at least has the honesty to build a parachute. This is where I want to introduce what I call the Human Ledger — the running account I keep in every analysis, the one that measures a protocol not by its total value locked but by the trust it asks its users to extend. On the Human Ledger, Avior scores higher than it has any right to. It asks for human approval before dangerous operations. It promises not to train on customer data. It offers three modes of autonomy, from assisted to fully autonomous, letting the operator choose how much control to surrender. These are the gestures of a system that understands it is handling something fragile. Now run the same ledger against the crypto protocols that spent 2024 promising autonomous agents — systems that would trade, rebalance, and govern on your behalf with no approval gate, no rollback, and no explanation. The comparison is unflattering. The enterprise AI industry, dismissed by crypto as the epitome of centralization, is building more governance into its agents than the decentralized projects that lecture the world about trust. The three operating modes are worth pausing on, because they are a taxonomy of decentralization that crypto should study. Assisted mode keeps the human in the loop for every action. Human-in-the-loop mode lets the agent act but requires approval for consequential decisions. Fully autonomous mode lets the agent act alone. Read that sequence again, and you will recognize it as the progressive decentralization roadmap that every DAO claims to follow and almost none completes. The difference is that Avior ships all three modes and lets the customer choose, while most DAOs ship a governance token and a promise that decentralization is coming. The token, as I have argued since the 2017 crisis, is a non-dividend share whose only hope is that a later buyer takes the bag. The three modes are a control surface. One is a product. The other is a hope. I keep coming back to the missing efficacy data, because it is the thread that ties enterprise AI to the worst instincts of crypto. SolarWinds published no adoption numbers, no measured improvements, no before-and-after. In the same week, DXC published 500 use cases and a 95 percent reduction in resolution time. The industry can prove its claims. SolarWinds declined to. This is the 2017 ICO problem, reborn in enterprise software. In 2017, I paused my technical consulting to audit the whitepapers of twenty-three prominent Ethereum tokens. Eighteen of them had no philosophical foundation, no community value proposition, nothing but speculation dressed as technology. The whitepaper was the product. The promise was the proof. What SolarWinds is doing now — shipping a governed agent with a governance narrative and no evidence — is the same move, executed by adults in a boardroom instead of anonymous developers on a forum. The difference is that enterprise buyers will eventually demand the data, because their procurement departments answer to shareholders. Crypto buyers never did. We bought the narrative, we bought the governance theater, and we called it a market. The competitive matrix tells its own story, and it is a story about the migration of value. Datadog, Dynatrace, ServiceNow, LogicMonitor — all of them shipped governed agents within the same window, and all of them, to varying degrees, have begun to publish evidence. The competition has moved from whose model is smarter to who can prove their agent is safe. This is the evidence competition, and it is the healthiest thing to happen to enterprise software in years. SolarWinds is the one player in the matrix with no data, and it is also the one with the deepest installed base — which means its strategy is to convert distribution into trust before anyone demands proof. In crypto, we ran this experiment in reverse. We let distribution precede proof, let the token precede the utility, let the narrative precede the audit. The result was a decade of projects that could not survive a single honest question. If Avior follows the same path, the configuration switch will flip for 300,000 customers and then flip back the moment the renewal invoices arrive without results to justify them. The land-and-expand model is the quietest and most revealing detail of all. SolarWinds is not selling Avior to strangers. It is flipping a configuration switch for 300,000 existing customers. Near-zero customer acquisition cost. Maximum adoption friction removed. If you have ever watched a DeFi protocol bootstrap its total value locked with liquidity mining rewards, you already understand this strategy. You subsidize the metric until the metric looks organic. The 300,000 customers are not a market; they are a distribution channel, and the agent's adoption curve will be a function of how easy the switch is to flip, not how good the product is. I spent the 2020 summer watching protocols inflate their TVL with emissions that vanished the moment the rewards stopped. The users were never there. The number was. SolarWinds has 300,000 customers who are already there — which is a genuine advantage, and also a genuine risk. If Avior's value is inseparable from the installed base, then the product is a bundling strategy, not a technology. And bundling strategies are judged by retention, not by capability. Which brings us back, always, to the data that was never published. The most consequential silence in the Avior release is the identity of the underlying model. Is it self-developed? Is it a third-party API? An open-source model? This is not a trivial detail. It is the difference between a company that owns its intelligence and a company that rents it. And it is the difference between a data flywheel that spins and one that stalls. Avior promises that customer data will not train the model — a promise that protects the customer and starves the product. If the model improves, it improves because someone else's data trained it. If it does not, the deterministic execution layer is a ceiling, not a foundation. I have watched crypto protocols make the same trade. They promise that your data stays yours, that your keys stay yours, that your sovereignty is absolute — and then they wonder why their networks never learn, never improve, never compound. Sovereignty has a cost. It is paid in feedback loops that never close. Avior is paying that cost in enterprise software, and the missing efficacy data may be the first visible symptom. In 2024, when the spot Bitcoin ETFs pulled more than fifty billion dollars of institutional capital into an asset designed to escape institutions, I wrote a guide called Institutional Entry, Individual Sovereignty. The thesis was simple: you can let the institutions in without letting them rewrite the rules, but only if you build the safeguards before the capital arrives, not after. Avior is the enterprise version of that thesis, and it is being tested in real time. The institutions — the CIOs, the compliance officers, the procurement departments — are arriving, and they are demanding governance. The question is whether the governance they get will protect the individual operator or merely indemnify the vendor. A rollback that restores the last known good configuration protects the system. It does not necessarily protect the person who was blamed for the failure. The Human Ledger does not balance itself. Someone has to decide whose trust is being kept, and the answer is written in the policy layer, not the code. I have seen this pattern before, in the summer of 2021, when I reviewed a hundred major NFT collections and found that most of them had no cultural substance, only ownership tokens dressed as community. I wrote a report called Soul-less Pixels, and the market ignored it, because the market was busy celebrating the speculation. What I learned then applies now. Automation without purpose is just efficiency in service of nothing. Avior's tasks — merging alerts, restoring configs, cleaning inventory — are purposeful, which is why they are the right tasks to automate. But the industry's rush to ship autonomous agents is not purposeful. It is a land grab, and land grabs produce a lot of territory that no one wants to govern. I want to name the thing that enterprise AI has rediscovered and crypto has forgotten. It is verifiable execution. For a decade, the blockchain industry has promised that we could prove what a machine did without trusting the machine. Zero-knowledge proofs, optimistic rollups, fraud proofs, on-chain governance — all of it was machinery for the same goal: a world where execution is verifiable, not merely asserted. We built that machinery, and then we buried it under speculation. Meanwhile, in the unglamorous back offices of IT operations, SolarWinds is shipping a deterministic execution layer that constrains an unreliable model, logs its trajectory, gates its dangerous actions, and rolls back its mistakes. It is not cryptographically verifiable. It is not trustless. But it is governed, and it is in production, and it is doing the work that crypto only ever theorized. The irony is almost too clean: the trust layer that the blockchain promised the world is being built by the company that the blockchain was supposed to make unnecessary. Here is the contrarian turn, and it is the one most crypto readers will resist. The instinct is to dismiss Avior as centralized, corporate, and therefore irrelevant — another walled garden, another surveillance tool. That instinct is a luxury we can no longer afford. The real danger in 2026 is not centralized AI. It is ungoverned AI, and the crypto industry is the worst offender in the room. We have spent three years shipping autonomous agents with no approval gates, no rollback, no policy layer — and we have dressed the absence of governance as a feature, calling it permissionless. We built towers of glass on beds of sand and told ourselves the glass was the point. Avior, for all its corporate baggage and its missing data, is at least asking the right question: how do we let a machine act without letting it run wild? Crypto answered that question in 2016 and got The DAO. Enterprise AI is answering it in 2026 with a control surface and a parachute. The uncomfortable possibility is that the governed machine will earn trust faster than the trustless one — because trust, in the end, is not a property of the code. It is a relationship. And relationships require accountability, not immutability. So watch the deterministic turn. Watch it spread from IT operations to finance, to healthcare, to the systems that decide whether a loan is approved or a diagnosis is trusted. Watch the governed machine become the default, and ask yourself, honestly, where the blockchain fits when the world has learned to constrain AI without it. Truth is not mined; it is revealed in the dark. The chain that cannot be governed may not be the chain that survives. Faith in code requires a heart for humanity — and the humans, this time, are building the guardrails themselves.