AI Agents Need Identity Before They Can Pay, Act, or Escalate

AI agents are getting closer to real-world authority, and that is exactly why identity is becoming the first problem, not the last one. Before an agent can buy, pay, file, or escalate, someone has to decide who it is, what it may do, and how that decision gets recorded.

The next phase of AI agents is not about making them more autonomous by default. It is about deciding which identities they can assume, which actions they can take, and how tightly each action is supervised.

The current reporting mix is unusually coherent: identity vendors are talking about delegated stacks, payment companies are piloting agentic flows, security teams are demanding enforcement, and standards discussions are starting to converge on authorization. That is how a market shift looks before it becomes a category.

The practical meaning of this story is that the industry is moving from novelty to operating discipline. Payments, support workflows, and enterprise automation are all moving toward delegated action and the stakes are whether agentic commerce becomes safe enough for production or chaotic enough to stall adoption are now in the same conversation, which tells you that capability alone no longer closes the sale.

What the reporting set is saying

Outlet	Headline	Signal
Biometric Update	Delegated identity stack emerges for AI agents to authenticate, act and pay	Shows the identity layer is being designed specifically for agentic action.
Tech Times	AI Agent Authorization Gets First Patented Three-Layer Trust Stack From Daon	Highlights that authorization is becoming a product race, not just a policy issue.
The Hacker News	Seeing AI Agents Is Not Enough. Security Teams Must Enforce What They Can Do	Makes least privilege the central security principle.
pymnts.com	48% of Online Shoppers Now Use AI Before Buying	Shows that AI is already sitting near purchase decisions.
Biometric Update	Keyfactor to acquire Cofide as AI agents drive identity shift	Shows the identity market adjusting to machine actors.
TNGlobal	Visa, China's Lianlian complete first live B2B agentic payment transaction using AI agent	Demonstrates that agentic payment flows are leaving the lab.
finews.asia	Visa and Lianlian Take Agentic AI Into B2B Payments	Confirms the payment stack is beginning to support delegated transactions.
TradingView	Nuvei Completes First-Party In-Agent Payment with Visa; Unveils Merchant-Led Agentic Payments Strategy	Shows the market experimenting with different payment authorization paths.
Microsoft	Microsoft Agent 365, now generally available, expands capabilities and integrations	Signals that enterprise governance for agents is being productized.
Biometric Update	Identity control plane emerges as next battleground for AI agents	Summarizes where the market is converging.

Biometric Update is useful here because delegated identity stack emerges for ai agents to authenticate, act and pay is not just a headline; it points to a specific market pressure. The story is less about any one announcement than about the fact that multiple observers are converging on the same conclusion. Shows the identity layer is being designed specifically for agentic action.

That convergence matters. When several sources keep circling the same pattern, the safest interpretation is that the ecosystem is adjusting to a new baseline. In this case, the baseline is that AI has to prove itself on cost, trust, and workflow fit instead of merely intelligence in isolation.

Tech Times is useful here because ai agent authorization gets first patented three-layer trust stack from daon is not just a headline; it points to a specific market pressure. The story is less about any one announcement than about the fact that multiple observers are converging on the same conclusion. Highlights that authorization is becoming a product race, not just a policy issue.

That convergence matters. When several sources keep circling the same pattern, the safest interpretation is that the ecosystem is adjusting to a new baseline. In this case, the baseline is that AI has to prove itself on cost, trust, and workflow fit instead of merely intelligence in isolation.

The Hacker News is useful here because seeing ai agents is not enough. security teams must enforce what they can do is not just a headline; it points to a specific market pressure. The story is less about any one announcement than about the fact that multiple observers are converging on the same conclusion. Makes least privilege the central security principle.

That convergence matters. When several sources keep circling the same pattern, the safest interpretation is that the ecosystem is adjusting to a new baseline. In this case, the baseline is that AI has to prove itself on cost, trust, and workflow fit instead of merely intelligence in isolation.

pymnts.com is useful here because 48% of online shoppers now use ai before buying is not just a headline; it points to a specific market pressure. The story is less about any one announcement than about the fact that multiple observers are converging on the same conclusion. Shows that AI is already sitting near purchase decisions.

That convergence matters. When several sources keep circling the same pattern, the safest interpretation is that the ecosystem is adjusting to a new baseline. In this case, the baseline is that AI has to prove itself on cost, trust, and workflow fit instead of merely intelligence in isolation.

Biometric Update is useful here because keyfactor to acquire cofide as ai agents drive identity shift is not just a headline; it points to a specific market pressure. The story is less about any one announcement than about the fact that multiple observers are converging on the same conclusion. Shows the identity market adjusting to machine actors.

That convergence matters. When several sources keep circling the same pattern, the safest interpretation is that the ecosystem is adjusting to a new baseline. In this case, the baseline is that AI has to prove itself on cost, trust, and workflow fit instead of merely intelligence in isolation.

TNGlobal is useful here because visa, china's lianlian complete first live b2b agentic payment transaction using ai agent is not just a headline; it points to a specific market pressure. The story is less about any one announcement than about the fact that multiple observers are converging on the same conclusion. Demonstrates that agentic payment flows are leaving the lab.

That convergence matters. When several sources keep circling the same pattern, the safest interpretation is that the ecosystem is adjusting to a new baseline. In this case, the baseline is that AI has to prove itself on cost, trust, and workflow fit instead of merely intelligence in isolation.

finews.asia is useful here because visa and lianlian take agentic ai into b2b payments is not just a headline; it points to a specific market pressure. The story is less about any one announcement than about the fact that multiple observers are converging on the same conclusion. Confirms the payment stack is beginning to support delegated transactions.

That convergence matters. When several sources keep circling the same pattern, the safest interpretation is that the ecosystem is adjusting to a new baseline. In this case, the baseline is that AI has to prove itself on cost, trust, and workflow fit instead of merely intelligence in isolation.

TradingView is useful here because nuvei completes first-party in-agent payment with visa; unveils merchant-led agentic payments strategy is not just a headline; it points to a specific market pressure. The story is less about any one announcement than about the fact that multiple observers are converging on the same conclusion. Shows the market experimenting with different payment authorization paths.

That convergence matters. When several sources keep circling the same pattern, the safest interpretation is that the ecosystem is adjusting to a new baseline. In this case, the baseline is that AI has to prove itself on cost, trust, and workflow fit instead of merely intelligence in isolation.

Microsoft is useful here because microsoft agent 365, now generally available, expands capabilities and integrations is not just a headline; it points to a specific market pressure. The story is less about any one announcement than about the fact that multiple observers are converging on the same conclusion. Signals that enterprise governance for agents is being productized.

That convergence matters. When several sources keep circling the same pattern, the safest interpretation is that the ecosystem is adjusting to a new baseline. In this case, the baseline is that AI has to prove itself on cost, trust, and workflow fit instead of merely intelligence in isolation.

Biometric Update is useful here because identity control plane emerges as next battleground for ai agents is not just a headline; it points to a specific market pressure. The story is less about any one announcement than about the fact that multiple observers are converging on the same conclusion. Summarizes where the market is converging.

That convergence matters. When several sources keep circling the same pattern, the safest interpretation is that the ecosystem is adjusting to a new baseline. In this case, the baseline is that AI has to prove itself on cost, trust, and workflow fit instead of merely intelligence in isolation.

The old assumption and the new reality

Old assumption	New reality	Why it matters
Treat an agent like a smarter chatbot	Treat an agent like a delegated identity with permissions	The security model changes completely.
Let the model decide how far it can go	Predefine action limits, approval steps, and audit logs	Autonomy becomes bounded rather than implied.
Assume payments are the main hard part	Assume identity and authorization are the hard part	Money follows trust, not the other way around.
Use generic enterprise controls	Design controls specifically for agentic behavior	The control plane has to match the new actor type.

The old assumption was treat an agent like a smarter chatbot. The new reality is treat an agent like a delegated identity with permissions. That shift sounds incremental, but it changes the business model underneath the product. Once the new reality takes hold, the vendor has to manage procurement, support, policy, and user expectations all at once.

The security model changes completely. That is what makes the story durable. It is not just a technical change. It is a change in how the product is justified inside an organization or a consumer ecosystem.

The old assumption was let the model decide how far it can go. The new reality is predefine action limits, approval steps, and audit logs. That shift sounds incremental, but it changes the business model underneath the product. Once the new reality takes hold, the vendor has to manage procurement, support, policy, and user expectations all at once.

Autonomy becomes bounded rather than implied. That is what makes the story durable. It is not just a technical change. It is a change in how the product is justified inside an organization or a consumer ecosystem.

The old assumption was assume payments are the main hard part. The new reality is assume identity and authorization are the hard part. That shift sounds incremental, but it changes the business model underneath the product. Once the new reality takes hold, the vendor has to manage procurement, support, policy, and user expectations all at once.

Money follows trust, not the other way around. That is what makes the story durable. It is not just a technical change. It is a change in how the product is justified inside an organization or a consumer ecosystem.

The old assumption was use generic enterprise controls. The new reality is design controls specifically for agentic behavior. That shift sounds incremental, but it changes the business model underneath the product. Once the new reality takes hold, the vendor has to manage procurement, support, policy, and user expectations all at once.

The control plane has to match the new actor type. That is what makes the story durable. It is not just a technical change. It is a change in how the product is justified inside an organization or a consumer ecosystem.

What this means for the market

AI Agents Need Identity Before They Can Pay, Act, or Escalate is easiest to understand as a systems story. The headline is useful, but the real shift is structural: the market is deciding whether AI should be judged by model quality, operating cost, and deployment friction at the same time. Once those variables are bundled together, the launch stops being a demo and starts becoming a procurement decision. The stakes are whether agentic commerce becomes safe enough for production or chaotic enough to stall adoption is the deeper business question. If the answer is yes, the AI layer turns into infrastructure. If the answer is no, it stays a pilot. That divide is what separates a headline from a platform.

That is why agentic identity and payments matters now. The industry is no longer asking only whether a model can do the task. It is asking whether the surrounding product can reduce the total cost of doing the task repeatedly, safely, and at scale. That sounds like a subtle change until the bill arrives in the form of compute spend, support overhead, or compliance risk. The reason these stories feel more consequential than a normal product refresh is that they all point to the same operating layer: who gets access, how actions are bounded, where liability lands, and how much of the workflow the model is allowed to touch. Those are not cosmetic questions. They are the conditions of adoption.

The current reporting set shows a market moving from symbolic capability toward measurable utility. The next phase of AI agents is not about making them more autonomous by default. It is about deciding which identities they can assume, which actions they can take, and how tightly each action is supervised. That sentence captures the real pressure on the vendor: buyers want results they can compare, managers want costs they can defend, and operators want workflows they can repeat without improvising every time. A lot of AI coverage still treats every release as if the main event were the intelligence itself. The better read is that the intelligence is now table stakes. The market is fighting over packaging, policy, permissioning, and the economics of repeated use. That is where differentiation now lives.

The economics matter because payments, support workflows, and enterprise automation are all moving toward delegated action. In practice, that means the winning product is not necessarily the one with the flashiest benchmark chart. It is the one that makes a real task cheaper to start, easier to supervise, and less expensive to correct when the model drifts. Agentic identity and payments also reveals how quickly AI has moved from optional tool to embedded dependency. Once a product sits between a person and a recurring job, the surrounding company has to care about reliability, defaults, logs, escalation paths, and cost controls. The software becomes part of the organization whether leadership wants that or not.

The stakes are whether agentic commerce becomes safe enough for production or chaotic enough to stall adoption is the deeper business question. If the answer is yes, the AI layer turns into infrastructure. If the answer is no, it stays a pilot. That divide is what separates a headline from a platform. That is why buyers have become more demanding. They are no longer impressed by a general claim that the model is smart. They want to know what it replaces, what it costs to run, how often it fails, and who gets paged when it does. Those are the questions that turn a launch into a durable market category.

The reason these stories feel more consequential than a normal product refresh is that they all point to the same operating layer: who gets access, how actions are bounded, where liability lands, and how much of the workflow the model is allowed to touch. Those are not cosmetic questions. They are the conditions of adoption. The strategic risk for the vendor is obvious. If the model is too expensive, the buyer limits use. If it is too permissive, security pushes back. If it is too restrictive, the workflow breaks. Every serious AI product now lives inside that triangle, and the company that manages it best wins the right to be considered default.

A lot of AI coverage still treats every release as if the main event were the intelligence itself. The better read is that the intelligence is now table stakes. The market is fighting over packaging, policy, permissioning, and the economics of repeated use. That is where differentiation now lives. Agentic identity and payments also changes how competitors behave. Once one company frames the category around cost, permissions, or boundaries, every rival has to answer the same questions. The market narrows around a new standard, and the old 'can it do the task?' debate gets replaced by 'can it do the task under real constraints?'

Agentic identity and payments also reveals how quickly AI has moved from optional tool to embedded dependency. Once a product sits between a person and a recurring job, the surrounding company has to care about reliability, defaults, logs, escalation paths, and cost controls. The software becomes part of the organization whether leadership wants that or not. For operators, the implication is simple but uncomfortable: AI is becoming an operational control surface, not a side feature. That means product teams, security teams, legal teams, and finance teams all care about the same system for different reasons. The launch lands successfully only if it satisfies all of them at once.

That is why buyers have become more demanding. They are no longer impressed by a general claim that the model is smart. They want to know what it replaces, what it costs to run, how often it fails, and who gets paged when it does. Those are the questions that turn a launch into a durable market category. The current reporting mix is unusually coherent: identity vendors are talking about delegated stacks, payment companies are piloting agentic flows, security teams are demanding enforcement, and standards discussions are starting to converge on authorization. That is how a market shift looks before it becomes a category. That context is what keeps the story from becoming generic. The point is not that AI is everywhere. The point is that the rules around AI are hardening fast enough to reshape who can use it, how, and at what price.

The strategic risk for the vendor is obvious. If the model is too expensive, the buyer limits use. If it is too permissive, security pushes back. If it is too restrictive, the workflow breaks. Every serious AI product now lives inside that triangle, and the company that manages it best wins the right to be considered default. AI Agents Need Identity Before They Can Pay, Act, or Escalate is easiest to understand as a systems story. The headline is useful, but the real shift is structural: the market is deciding whether AI should be judged by model quality, operating cost, and deployment friction at the same time. Once those variables are bundled together, the launch stops being a demo and starts becoming a procurement decision.

Agentic identity and payments also changes how competitors behave. Once one company frames the category around cost, permissions, or boundaries, every rival has to answer the same questions. The market narrows around a new standard, and the old 'can it do the task?' debate gets replaced by 'can it do the task under real constraints?' That is why agentic identity and payments matters now. The industry is no longer asking only whether a model can do the task. It is asking whether the surrounding product can reduce the total cost of doing the task repeatedly, safely, and at scale. That sounds like a subtle change until the bill arrives in the form of compute spend, support overhead, or compliance risk.

For operators, the implication is simple but uncomfortable: AI is becoming an operational control surface, not a side feature. That means product teams, security teams, legal teams, and finance teams all care about the same system for different reasons. The launch lands successfully only if it satisfies all of them at once. The current reporting set shows a market moving from symbolic capability toward measurable utility. The next phase of AI agents is not about making them more autonomous by default. It is about deciding which identities they can assume, which actions they can take, and how tightly each action is supervised. That sentence captures the real pressure on the vendor: buyers want results they can compare, managers want costs they can defend, and operators want workflows they can repeat without improvising every time.

The current reporting mix is unusually coherent: identity vendors are talking about delegated stacks, payment companies are piloting agentic flows, security teams are demanding enforcement, and standards discussions are starting to converge on authorization. That is how a market shift looks before it becomes a category. That context is what keeps the story from becoming generic. The point is not that AI is everywhere. The point is that the rules around AI are hardening fast enough to reshape who can use it, how, and at what price. The economics matter because payments, support workflows, and enterprise automation are all moving toward delegated action. In practice, that means the winning product is not necessarily the one with the flashiest benchmark chart. It is the one that makes a real task cheaper to start, easier to supervise, and less expensive to correct when the model drifts.

AI Agents Need Identity Before They Can Pay, Act, or Escalate is easiest to understand as a systems story. The headline is useful, but the real shift is structural: the market is deciding whether AI should be judged by model quality, operating cost, and deployment friction at the same time. Once those variables are bundled together, the launch stops being a demo and starts becoming a procurement decision. The stakes are whether agentic commerce becomes safe enough for production or chaotic enough to stall adoption is the deeper business question. If the answer is yes, the AI layer turns into infrastructure. If the answer is no, it stays a pilot. That divide is what separates a headline from a platform.

That is why agentic identity and payments matters now. The industry is no longer asking only whether a model can do the task. It is asking whether the surrounding product can reduce the total cost of doing the task repeatedly, safely, and at scale. That sounds like a subtle change until the bill arrives in the form of compute spend, support overhead, or compliance risk. The reason these stories feel more consequential than a normal product refresh is that they all point to the same operating layer: who gets access, how actions are bounded, where liability lands, and how much of the workflow the model is allowed to touch. Those are not cosmetic questions. They are the conditions of adoption.

The current reporting set shows a market moving from symbolic capability toward measurable utility. The next phase of AI agents is not about making them more autonomous by default. It is about deciding which identities they can assume, which actions they can take, and how tightly each action is supervised. That sentence captures the real pressure on the vendor: buyers want results they can compare, managers want costs they can defend, and operators want workflows they can repeat without improvising every time. A lot of AI coverage still treats every release as if the main event were the intelligence itself. The better read is that the intelligence is now table stakes. The market is fighting over packaging, policy, permissioning, and the economics of repeated use. That is where differentiation now lives.

The economics matter because payments, support workflows, and enterprise automation are all moving toward delegated action. In practice, that means the winning product is not necessarily the one with the flashiest benchmark chart. It is the one that makes a real task cheaper to start, easier to supervise, and less expensive to correct when the model drifts. Agentic identity and payments also reveals how quickly AI has moved from optional tool to embedded dependency. Once a product sits between a person and a recurring job, the surrounding company has to care about reliability, defaults, logs, escalation paths, and cost controls. The software becomes part of the organization whether leadership wants that or not.

The stakes are whether agentic commerce becomes safe enough for production or chaotic enough to stall adoption is the deeper business question. If the answer is yes, the AI layer turns into infrastructure. If the answer is no, it stays a pilot. That divide is what separates a headline from a platform. That is why buyers have become more demanding. They are no longer impressed by a general claim that the model is smart. They want to know what it replaces, what it costs to run, how often it fails, and who gets paged when it does. Those are the questions that turn a launch into a durable market category.

The reason these stories feel more consequential than a normal product refresh is that they all point to the same operating layer: who gets access, how actions are bounded, where liability lands, and how much of the workflow the model is allowed to touch. Those are not cosmetic questions. They are the conditions of adoption. The strategic risk for the vendor is obvious. If the model is too expensive, the buyer limits use. If it is too permissive, security pushes back. If it is too restrictive, the workflow breaks. Every serious AI product now lives inside that triangle, and the company that manages it best wins the right to be considered default.

A lot of AI coverage still treats every release as if the main event were the intelligence itself. The better read is that the intelligence is now table stakes. The market is fighting over packaging, policy, permissioning, and the economics of repeated use. That is where differentiation now lives. Agentic identity and payments also changes how competitors behave. Once one company frames the category around cost, permissions, or boundaries, every rival has to answer the same questions. The market narrows around a new standard, and the old 'can it do the task?' debate gets replaced by 'can it do the task under real constraints?'

Agentic identity and payments also reveals how quickly AI has moved from optional tool to embedded dependency. Once a product sits between a person and a recurring job, the surrounding company has to care about reliability, defaults, logs, escalation paths, and cost controls. The software becomes part of the organization whether leadership wants that or not. For operators, the implication is simple but uncomfortable: AI is becoming an operational control surface, not a side feature. That means product teams, security teams, legal teams, and finance teams all care about the same system for different reasons. The launch lands successfully only if it satisfies all of them at once.

That is why buyers have become more demanding. They are no longer impressed by a general claim that the model is smart. They want to know what it replaces, what it costs to run, how often it fails, and who gets paged when it does. Those are the questions that turn a launch into a durable market category. The current reporting mix is unusually coherent: identity vendors are talking about delegated stacks, payment companies are piloting agentic flows, security teams are demanding enforcement, and standards discussions are starting to converge on authorization. That is how a market shift looks before it becomes a category. That context is what keeps the story from becoming generic. The point is not that AI is everywhere. The point is that the rules around AI are hardening fast enough to reshape who can use it, how, and at what price.

The strategic risk for the vendor is obvious. If the model is too expensive, the buyer limits use. If it is too permissive, security pushes back. If it is too restrictive, the workflow breaks. Every serious AI product now lives inside that triangle, and the company that manages it best wins the right to be considered default. AI Agents Need Identity Before They Can Pay, Act, or Escalate is easiest to understand as a systems story. The headline is useful, but the real shift is structural: the market is deciding whether AI should be judged by model quality, operating cost, and deployment friction at the same time. Once those variables are bundled together, the launch stops being a demo and starts becoming a procurement decision.

Agentic identity and payments also changes how competitors behave. Once one company frames the category around cost, permissions, or boundaries, every rival has to answer the same questions. The market narrows around a new standard, and the old 'can it do the task?' debate gets replaced by 'can it do the task under real constraints?' That is why agentic identity and payments matters now. The industry is no longer asking only whether a model can do the task. It is asking whether the surrounding product can reduce the total cost of doing the task repeatedly, safely, and at scale. That sounds like a subtle change until the bill arrives in the form of compute spend, support overhead, or compliance risk.

For operators, the implication is simple but uncomfortable: AI is becoming an operational control surface, not a side feature. That means product teams, security teams, legal teams, and finance teams all care about the same system for different reasons. The launch lands successfully only if it satisfies all of them at once. The current reporting set shows a market moving from symbolic capability toward measurable utility. The next phase of AI agents is not about making them more autonomous by default. It is about deciding which identities they can assume, which actions they can take, and how tightly each action is supervised. That sentence captures the real pressure on the vendor: buyers want results they can compare, managers want costs they can defend, and operators want workflows they can repeat without improvising every time.

The current reporting mix is unusually coherent: identity vendors are talking about delegated stacks, payment companies are piloting agentic flows, security teams are demanding enforcement, and standards discussions are starting to converge on authorization. That is how a market shift looks before it becomes a category. That context is what keeps the story from becoming generic. The point is not that AI is everywhere. The point is that the rules around AI are hardening fast enough to reshape who can use it, how, and at what price. The economics matter because payments, support workflows, and enterprise automation are all moving toward delegated action. In practice, that means the winning product is not necessarily the one with the flashiest benchmark chart. It is the one that makes a real task cheaper to start, easier to supervise, and less expensive to correct when the model drifts.

Scenarios to watch

Scenario	What happens	What to watch
Identity vendors win the trust layer	Agent authentication, authorization, and audit become standardized	Watch delegated identity frameworks and payment pilots.
Security teams stay skeptical	Agentic commerce advances slowly and under tight limits	Watch approval workflows, audit tooling, and policy enforcement.
Consumers accept bounded autonomy	Buy, pay, and escalate actions become normal for low-risk tasks	Watch how clearly users can see the agent's authority boundaries.

If identity vendors win the trust layer, then agent authentication, authorization, and audit become standardized. That is the difference between a launch cycle and a durable category shift. The first produces a spike in attention; the second changes how teams budget, approve, and deploy the product every day.

What to watch next is watch delegated identity frameworks and payment pilots.. That is where the story will either compound or slow down. The market does not reward clever framing for long if the operational evidence fails to show up.

If security teams stay skeptical, then agentic commerce advances slowly and under tight limits. That is the difference between a launch cycle and a durable category shift. The first produces a spike in attention; the second changes how teams budget, approve, and deploy the product every day.

What to watch next is watch approval workflows, audit tooling, and policy enforcement.. That is where the story will either compound or slow down. The market does not reward clever framing for long if the operational evidence fails to show up.

If consumers accept bounded autonomy, then buy, pay, and escalate actions become normal for low-risk tasks. That is the difference between a launch cycle and a durable category shift. The first produces a spike in attention; the second changes how teams budget, approve, and deploy the product every day.

What to watch next is watch how clearly users can see the agent's authority boundaries.. That is where the story will either compound or slow down. The market does not reward clever framing for long if the operational evidence fails to show up.

flowchart TD
    A[Human or enterprise identity] --> B[Delegated agent identity]
    B --> C[Permissioned action]
    C --> D[Payment, escalation, or workflow step]
    D --> E[Audit trail and review]

The bottom line

The stakes are whether agentic commerce becomes safe enough for production or chaotic enough to stall adoption is the real test, not whether the model can impress in a demo. The important question is whether the system can absorb the new behavior without passing hidden costs to the user, the buyer, or the public. That is the moment AI stops being a product story and becomes an operating model.

Agentic identity and payments is therefore less about the current headline than the next default. The companies that understand that shift will look more durable because they are selling control, trust, and repeatability. The ones that do not will keep discovering that the hard part of AI was never the answer; it was everything around it.