Manufacturing took 60 years to show these signals. Software took 20. AI is running the same pattern in about 4.
Every domain that gets industrialized passes through the same transformation. The signals are consistent enough that you can use them as a diagnostic: count how many are present in a given industry, and you have a reasonable estimate of how far along the collapse is. Zero signals means the domain is stable. Four or five means disruption is already underway and the incumbents are negotiating their terms. Seven means the domain has already collapsed and you're watching the cleanup.
Right now, several industries are sitting at 3-4 signals simultaneously. That's the inflection point. Here's what to watch for.
The Pattern Behind the Pattern
Before getting into the signatures, it's worth understanding why they appear at all. When a domain industrializes, the core shift is that tacit knowledge gets made explicit. Things that previously required a person's judgment get encoded into procedures, then into software, then into automated systems. Each step in that encoding produces one of these signatures as a byproduct.
This isn't a new observation. Frederick Taylor made it about factory floors in 1911. W. Edwards Deming made it about quality control in the 1950s. What's new is the speed. AI compresses the encoding timeline dramatically because language models already carry a lot of the tacit knowledge as latent structure. You don't have to watch an expert for 10,000 hours to extract their heuristics. You train on the corpus of what experts wrote down, and you get most of the way there in weeks.
The SolveEverything framework maps this process systematically across domains. The 7 signatures below are the observable footprints it leaves behind.
Signature 1: Payment Shifts from Effort to Outcome
The clearest early signal is pricing model change. When a domain is labor-intensive and outcomes are hard to measure, clients pay for time. Hourly rates, monthly retainers, project fees. When the work becomes automatable and outcomes become measurable, the pricing inverts. Clients pay for results, and whoever bears the execution risk is whoever has the most confidence in the system.
You can see this in customer service right now. Traditional call centers bill per agent-hour. Freebot, which fights corporate customer service on behalf of consumers, charges per resolution. It only collects when it wins. That's a statement about confidence in the system's reliability, not just a marketing decision.
The same shift is happening in e-commerce. Freway's checkout recovery agent Janine runs on commission-only pricing, collecting only on recovered sales. A human sales team working on pure commission would be unusual. An AI agent doing it is the natural model, because the marginal cost of an additional attempt is near zero.
OneShot's pricing for agent tools follows the same logic: agents pay per successful API call, per verified action, per completed task. The infrastructure cost is low enough that outcome-based pricing is viable at scale.
When you see a domain's incumbents defending hourly billing against outcome-based challengers, that's the first signature appearing. The incumbents aren't wrong that their time is valuable. They're wrong that clients will keep paying for it once the alternative exists.
Signature 2: Documents Become Machine-Verifiable Proofs
The second signal is subtler but more structurally important. In every pre-industrial domain, the output of work is a document: a report, a proposal, an analysis, an invoice. Documents are human-readable artifacts that require trust to interpret. You believe the consultant's analysis because you trust the consultant, not because you can independently verify the methodology in real time.
When a domain industrializes, documents get replaced by receipts. A receipt is a machine-verifiable proof that something happened: a transaction was completed, a call was made, a message was delivered. Receipts don't require trust. They require cryptographic verification or API confirmation.
In agent commerce, this is already happening. When an OneShot agent makes a voice call, sends an email, or runs a verification, the output isn't a report saying "the call was made." It's a structured payload with timestamps, confirmation codes, and verifiable status. The x402 protocol that handles micropayments between agents is the same idea applied to money: payment receipts that don't require a trusted intermediary to confirm.
Legal and accounting are deep in Signature 2 right now. The deliverable used to be a memo. It's becoming a structured data output that feeds directly into downstream systems. The memo still exists as a human-readable summary, but it's no longer the primary artifact. The machine-readable payload is.
Signature 3: Discrete Projects Become Continuous Pipelines
Pre-industrial work is episodic. A client has a problem, they hire someone, the project runs for a defined period, it ends. The engagement model is project-shaped.
Industrial work is continuous. The factory doesn't stop between orders. The monitoring system doesn't take weekends off. When AI enters a domain, the work shifts from episodic to always-on, because the marginal cost of continuous operation is negligible compared to the cost of starting and stopping.
Freway's checkout intervention is a clean example. A human sales team can't watch every cart in real time and intervene at the exact moment hesitation appears. An AI agent can, because it costs essentially nothing to have it running continuously. The business model only makes sense as a pipeline, not a project.
In recruiting, the old model was "open a req, run a search, close the req." The emerging model is continuous candidate pipeline management where the agent is always sourcing, always screening, always maintaining warm relationships with candidates who aren't ready yet. The episodic project structure made sense when human attention was the scarce resource. It stops making sense when the scarce resource is API calls.
Signature 4: Individual Heroics Yield to Systems Engineering
Every pre-industrial domain has its version of the 10x performer. The rainmaker partner at the law firm. The senior engineer who knows where all the bodies are buried. The account executive who can close deals nobody else can. These people are genuinely valuable, and organizations hoard them.
When a domain industrializes, the 10x performer's edge shrinks. Their tacit knowledge gets encoded into the system. Their judgment calls become decision trees. Their relationships get replaced by automated touchpoints that are more consistent if less personal.
This doesn't mean expertise becomes worthless. It means expertise shifts from execution to system design. The valuable person stops being the one who does the work and starts being the one who designs the system that does the work. That's a different skill set, and most 10x performers aren't trained for it.
In software development, this is already visible. The gap between a senior and junior engineer used to be 10x in output. With good AI tooling, a junior engineer with strong systems thinking can produce more than a senior engineer who's optimized for individual coding speed. The bottleneck moved from coding to architecture, from execution to orchestration.
Agent commerce is at the early stages of this shift. The valuable people right now are the ones who can design agent workflows, define verification criteria, and set up the observability to know when the system is failing. Raw prompt engineering is already commoditizing.
Signature 5: Proprietary Secrecy Gives Way to Ecosystem Transparency
Incumbents in pre-industrial domains protect their methods. The secret sauce is the moat. Consulting firms don't publish their frameworks. Law firms don't share their playbooks. The information asymmetry between expert and client is the business model.
When a domain industrializes, the moat moves. The winners aren't the ones with the best-kept secrets. They're the ones who established the benchmarks that everyone else gets measured against. Open standards, public APIs, shared evaluation criteria. The company that writes the scoreboard controls the game.
You can see this in AI benchmarks right now. The organizations publishing evaluation frameworks (MMLU, HumanEval, LMSYS Arena) have more influence over the direction of AI development than most of the companies building closed models. They set the terms of comparison. Everyone else optimizes to their metrics.
Soul.Markets is betting on this dynamic in agent commerce. The soul.md identity format for agents is public. The marketplace is open. The thesis is that whoever establishes the standard for how agents describe their capabilities and pricing will have more durable influence than whoever builds the best proprietary agent. The format becomes the moat, not the secret.
Signature 6: Average-Case Optimization Shifts to Tail-Risk
Pre-industrial quality control focuses on the median. Is the typical output good enough? Are most clients satisfied? This makes sense when you're managing human workers, because you can't observe every output and you have to sample.
Industrial quality control focuses on the tail. What's the worst case? What happens when the system fails? What's the failure mode distribution? This matters more when the system is running at scale, because a 0.1% failure rate that's fine for a human team becomes catastrophic when you're processing a million transactions.
In agent commerce, this shift is happening fast. When Freebot makes a hundred calls on a client's behalf, the median call quality matters less than the worst call. If one call in a hundred does something harmful (makes a false claim, agrees to something it shouldn't), the reputational damage outweighs the value of the other 99 successful calls.
This is why verification is a first-class tool in OneShot's SDK. The tools aren't just about execution. They're about confirming that execution met the criteria before counting it as success. Tail-risk optimization requires observability at the level of individual actions, not just aggregate outcomes.
In financial services, this shift is already complete. Risk models focus on value-at-risk, stress tests, worst-case scenarios. The average case is assumed to be fine. The regulation and the engineering effort concentrate on the tail. AI-native industries will get there faster than financial services did, because the failure modes are more visible and the stakes become clear earlier.
Signature 7: Talent Hoarding Becomes Compute Liquidity
The final signature is the most structurally disruptive. Pre-industrial organizations compete for people. The war for talent is real because talent is genuinely scarce and genuinely differentiated. You hire the best people you can, you keep them, you build institutional knowledge.
When a domain industrializes, the scarce resource shifts from people to compute. You don't hire more analysts. You allocate more tokens. You don't retain the best researcher. You fine-tune the model on the best researcher's outputs and run it at 1000x scale. The organization that can deploy compute liquidity faster than competitors wins, regardless of headcount.
This is already playing out in financial research. The banks with the most analysts aren't producing the most alpha. The banks with the best data infrastructure and the fastest model iteration cycles are. The talent war is real, but the talent being fought over is ML engineers and data architects, not subject matter experts.
In agent commerce, compute liquidity means the ability to spin up capacity instantly when demand spikes and wind it down when it doesn't. Freway's checkout agent doesn't need to hire more Janines when a merchant runs a flash sale. It scales the compute allocation. The cost structure is fundamentally different from a human team, and the operational model has to be different too.
How to Count the Signals
Run through this diagnostic for any domain you're evaluating:
- Are there outcome-based pricing models competing with effort-based ones?
- Are the primary deliverables shifting from documents to machine-readable data?
- Are engagements becoming continuous rather than episodic?
- Is systems design becoming more valuable than individual execution?
- Are the winners publishing standards rather than hoarding methods?
- Is quality control shifting from median to tail-risk?
- Is headcount growth decoupling from output growth?
Three signals present means disruption is early-stage. Incumbents are dismissing it. Four or five means the disruption is real and the incumbents are starting to respond, usually by acquiring or copying the challengers. Six or seven means the domain has already structurally changed and you're watching the institutional lag catch up.
Customer service is at 5-6 right now. Legal research is at 3-4. Accounting is at 4-5. Medical diagnosis is at 2-3 (regulatory friction is slowing signals 1 and 3). Software development is at 6.
The Prediction
By Q2 2027, outcome-based pricing will be the majority pricing model (by transaction count, not revenue) in at least three of these domains: customer service, legal research, financial analysis, and recruiting. The revenue majority will lag by 12-18 months because large enterprise contracts are slow to renegotiate, but the transaction majority will flip first in the mid-market.
The companies that survive that transition will be the ones that recognized Signature 5 earliest: they published the benchmarks, set the evaluation criteria, and became the scoreboard rather than just another player. Everyone else will be optimizing for metrics they didn't define.
If you're building in agent commerce and want to see what the infrastructure layer looks like at Signature 3-4, the OneShot documentation is a reasonable place to start. The patterns for how agents pay for tools, verify outcomes, and chain actions together are the same patterns that will run at scale when the domain fully collapses.