It's the End of Software That Cannot Adapt
TL:DR
A thoughtful post from Maddison Gay on August 6, 2026 helped spark this idea: AI agents won’t destroy software, but they will expose software that can’t adapt.
AI agents won’t eliminate operational software. They’ll reveal which systems were built to support change and which were built to resist it.
For the aviation aftermarket, that difference may determine which companies turn AI into real capacity and which remain trapped behind yesterday’s software.
A Gartner forecast suggested that as much as $234 billion in enterprise application software spending could be at risk from agentic AI by 2030.
Some call the coming disruption the “SaaSpocalypse.”
It’s an effective headline, but it points toward the wrong conclusion.
Software isn’t disappearing. Aviation companies will still need trusted systems to manage inventory, financial records, quality controls, repair histories, permissions, documentation and transactional truth.
An AI agent can’t simply become all of those things.
What is disappearing is the assumption that companies must keep reshaping their operations around inflexible software.
For years, that compromise was accepted as the cost of doing business. The ERP was installed, its workflow became the company’s workflow, and employees learned to compensate for everything it couldn’t do.
They created spreadsheets, watched shared inboxes, copied information between portals and kept critical knowledge in their heads.
When the software didn’t fit the operation, the operation was forced to fit the software.
AI agents challenge that arrangement because they shift the value of software away from how many screens an employee can navigate and toward how much work the system can help complete.
That’s the real disruption.
We Trained the Operation to Tolerate the Software
Enterprise software was meant to organize complex businesses. It created common records, standardized transactions and helped departments coordinate.
Over time, the relationship often reversed.
Instead of asking whether the software understood the operation, companies trained the operation to tolerate the software.
Sales teams quoted in the sequence the ERP required. Purchasing built side processes for vendor communication. Quality stored documents outside the records they supported. Managers relied on yesterday’s reports because the system could describe the past but couldn’t help coordinate what happened next.
These workarounds weren’t necessarily signs of poor management. They were rational responses to rigid systems.
Customers still had to be served. Parts had to move. Repairs had to be completed. Documentation had to be collected.
So people built an informal operating system around the official one.
Across the aviation aftermarket, the ERP may be the system of record, but the real operation is spread across email, spreadsheets, portals, shared drives, meetings and institutional memory.
That gap matters more now.
An experienced employee may recognize a hidden customer requirement, remember a supplier’s documentation problems or know that an apparent stock match is already committed elsewhere.
For AI to act responsibly, those relationships must exist in accessible data, rules and workflows.
Otherwise, the technology inherits the same blind spots as the system beneath it.
What Agents Actually Change
Traditional software waits for a user to perform a defined action. Someone logs in, finds a record, selects an option and moves the transaction forward.
An agent can recognize an event, gather context, apply rules, evaluate possible actions and initiate an approved next step.
Consider an incoming aviation RFQ.
A properly designed agent could identify the customer and requested part, determine urgency, review transaction history, evaluate inventory, inspect pricing rules, identify alternates, check documentation requirements and prepare the next action.
It might draft a quote, request vendor pricing, route an exception for approval or alert a salesperson that judgment is required.
That’s different from placing a chatbot over a database.
A chatbot can answer a question about a transaction. An operational agent may help advance it.
But only when the system can provide a trustworthy version of reality.
Agents still need clean data, visible relationships, reliable permissions, secure action layers, clear rules and defined escalation paths.
Without those foundations, an agent may act quickly while still acting incorrectly. It may use outdated inventory, overlook documentation, rely on a weak supplier record, bypass an approval or duplicate work happening elsewhere.
The AI may be capable.
The operating environment may not be.
That’s why so many agentic AI projects struggle before reaching production. Companies focus on the visible intelligence layer while leaving the underlying data, workflow and governance unresolved.
A polished interface can’t compensate for a fragmented operation.
Why Aviation Raises the Stakes
Every industry must consider accuracy, security and governance. Aviation adds another level of consequence.
The aftermarket doesn’t simply move products. It moves serialized components, trace documents, release certificates, condition information, maintenance records, quality approvals, export requirements and financial obligations.
A part number and price aren’t enough.
A component may physically exist and still be unusable. Its documentation may not support its identity. Its condition may not match the quote. Its certification may be missing or expired. Its trace may not satisfy the customer. Its shelf life may make it unsuitable.
The physical part and the evidence supporting it form one complete package.
An aviation agent can’t be judged only by how quickly it locates stock, drafts a response or moves information. It must understand the commercial, quality and compliance context surrounding the transaction.
That makes shallow automation especially risky.
A faster incorrect action isn’t an improvement.
Yet aviation is also an ideal environment for governed intelligence. The work is repetitive, contextual and document-heavy. Companies already hold years of customer history, supplier performance, quoting behavior, purchasing patterns, repair records and institutional knowledge.
The opportunity is substantial, provided intelligence strengthens the workflow rather than bypassing it.
The ERP Is Often the Real Constraint
Many aviation companies want AI. They see the potential and can identify processes consuming too much employee time.
Yet they often don’t know where to begin.
The limiting factor is frequently the ERP beneath the operation.
The company may own years of valuable data, but it’s divided across modules, exports, attachments and outside systems. The official workflow may live in the ERP while the practical workflow depends on email and personal knowledge.
Even accessing the company’s own information can become a project.
Before building an agent-driven process, a business may need to negotiate API access, purchase another integration layer, map proprietary fields or determine whether the system can safely execute an action at all.
At that point, the cost of automation is no longer primarily the intelligence.
It’s the effort required to overcome the software.
This is where the SaaSpocalypse argument becomes credible.
The software most at risk isn’t necessarily the platform with the fewest AI features. It’s the platform that makes useful automation unnecessarily difficult.
A system becomes vulnerable when it can’t expose its data, represent the real workflow, enforce controls or support a new application without a lengthy customization project.
An AI label won’t solve that.
Flexibility Still Requires Control
A flexible platform shouldn’t allow every user, application or agent to change the operation without restraint.
Aviation can’t accept that version of flexibility.
A company should be able to define which customers receive automated pricing, which transactions require approval, which documents must be present, which inventory can be allocated and which exceptions must be escalated.
It should also establish user roles, financial limits, quality holds, customer-specific requirements and boundaries on agent authority.
The goal isn’t to remove people from every decision.
It’s to remove unnecessary human effort from decisions the company has already learned how to make reliably.
Intelligence without control creates risk. Control without flexibility creates stagnation.
Aviation needs both.
From Added AI to Embedded Intelligence
There’s a meaningful difference between adding AI to a product and building a system that supports intelligence inside the operation.
Added AI often begins at the interface. It summarizes records, reads exported information, answers questions or produces recommendations.
Those capabilities may help, but they don’t necessarily change how work gets completed. The intelligence remains outside the transactional core.
Embedded intelligence begins with connected records, documents, permissions, events and actions.
It understands not only that a record exists, but how that record relates to a customer, transaction, approval, document and next step.
It can recognize a change, determine what should happen, act within defined authority, preserve what it did and route uncertainty to a person.
That’s the direction we believe operational software must take.
At ERP.Aero, the goal isn’t to attach intelligence to the outside of an aviation ERP and call the work complete. It’s to make intelligence part of the environment where quoting, inventory, purchasing, repairs, quality, documentation, finance and reporting already meet.
​ELIA, our Embedded Layer of Intelligence in Aviation, is part of that direction.
Its purpose isn’t simply to answer generic questions. It’s intended to work with the context already inside the operation, including RFQs, customer and vendor history, inventory, alternates, documents, certifications, assignments and exceptions.
The larger principle matters more than any individual feature.
Intelligence should know where it is, what the company allows and what operational evidence supports the action.
Software Should Become a Foundation for Building
No software company can predict every workflow its customers will need.
That’s especially true in aviation, where two companies serving the same market may maintain entirely different commercial rules, repair processes, customer requirements, approval structures and specialties.
Historically, businesses had three imperfect choices:
Accept the standard workflow.
Pay for customization.
Or build another process outside the ERP.
A platform model creates a fourth.
ERP.Aero Apps and the initiative at erp.aero/apps are being developed around the idea that aviation companies should be able to extend their operational foundation without rebuilding it.
The ERP provides trusted records, permissions, controls, business objects and transactional history. Applications extend that foundation for a specific workflow, customer need or market opportunity.
A company might build specialized quoting logic, a repair workflow, a teardown application, a compliance utility, a vendor collaboration tool, a customer portal, a logistics process or a new form of operational intelligence.
Developers and industry specialists could work from an aviation-native foundation instead of recreating the underlying ERP logic each time.
This also makes agents more practical.
A specialized agent shouldn’t have to rediscover what an RFQ, purchase order, work order, certification, quality hold or serialized part means. Those concepts should already exist within the platform.
The application or agent can focus on the differentiated workflow because the operational foundation is already there.
The future of ERP may not be one software company attempting to predict and own every feature a customer will ever need.
It may be a trusted foundation upon which customers, partners and an industry continue building.
The New Role of the ERP
This doesn’t make the ERP less important.
It makes it more foundational and less visible.
The ERP of the future must remain a system of record, but it must also become a system of control, workflow engine, permissions layer, development platform and intelligence environment.
It must protect consistency without forcing sameness.
The core platform should define the trusted structure of the operation. Applications should extend how individual companies work. Intelligence should help coordinate decisions and actions within the boundaries the business establishes.
AI doesn’t replace the operational foundation. It changes what that foundation must support.
Companies will still need reliable systems for inventory, accounting, purchasing, quality and repair history. They’ll simply expect those systems to be accessible, adaptable and able to participate in intelligent workflows.
Platforms that can’t make that transition may remain installed for years.
But customers will increasingly spend around them.
They’ll buy automation tools, external databases, integration platforms, intelligence products and custom development to accomplish what the core system can’t support.
That may be the real SaaSpocalypse: not the sudden extinction of software, but the gradual transfer of value away from the system that once sat at the center of the operation.
Eight Questions Aviation Leaders Should Ask
Before investing heavily in agents, aviation companies should assess whether their operating foundation is ready.
1. Can we trust the data?​
Are inventory, pricing, customer, supplier and documentation records complete and current enough to support automated action?
2. Does the system represent the real workflow?​
Or does the practical process depend on email, spreadsheets, meetings and employee memory?
3. Can approved applications access what they need?​
Does the platform provide usable APIs and action layers, or does every integration become a custom project?
4. Are authority and approvals clearly defined?​
Can the system determine what an agent may do, what requires human review and who owns each exception?
5. Are documents connected to the transactions they support?​
Can the system evaluate the evidence surrounding a part, quote, purchase, repair or release?
6. Can every automated action be reconstructed?​
Will the company know what happened, why it happened, which information was used and what authorized it?
7. Can the platform support what comes next?​
Can customers and developers build new workflows and applications without replacing the operational core?
8. Is the AI amplifying good, or making up for shortcomings? Meaning, is it solving for an architectural problem? And what would happen if the layer is removed?
These questions may reveal that the first AI project shouldn’t begin with an agent.
It may begin with data, workflows and governance.
That isn’t a delay in the AI strategy.
It’s the beginning of a credible one.
The Opportunity Is Larger Than Automation
The aviation aftermarket shouldn’t pursue AI simply because other industries are doing it.
It should use this moment to reconsider how operational systems support people, knowledge and execution.
The opportunity extends beyond faster quoting, better search and automated data entry.
Well-designed intelligence can preserve institutional knowledge as experienced employees retire. It can give newer team members access to customer and supplier history. It can identify missing information before a transaction progresses and connect commercial decisions with quality and operational consequences.
It can reduce repetitive work while strengthening consistency.
Most importantly, it can allow experienced people to spend more time applying judgment and less time transporting information between systems.
The goal isn’t to build an operation without people.
It’s to build one where people no longer compensate for the limitations of their software.
The Real “Aha” Isn’t About AI
The SaaSpocalypse won’t arrive as a single moment when software disappears.
It will unfold as businesses reconsider where software creates value.
They’ll become less willing to pay for interfaces that add work, closed systems that restrict access to their own information and rigid workflows that prevent the operation from evolving.
They’ll expect software to hold trusted records, enforce governance, support applications and provide a safe environment for intelligence.
Some platforms will adapt.
Others will add a chat window, rename a few features and leave the underlying experience unchanged.
But here is the more important realization:
AI agents may not replace the ERP. They may reveal what the ERP was always supposed to become.
Not another screen employees must learn.
Not another database they must work around.
A living operational foundation that understands the business, protects its rules, preserves its knowledge and helps move work forward.
That’s the real “aha.”
The future isn’t software disappearing into AI.
It’s software finally becoming useful enough to disappear into the operation.
The question is no longer whether intelligence will become part of aviation.
It already has.
The question is whether the system beneath it is helping the business build what comes next, or quietly standing in the way.
-About ERP.Aero
​ERP.Aero is an aviation-first ERP platform built to connect RFQs, quoting, inventory, purchasing, repairs, work orders, compliance, shipping, invoicing, reporting, and customer intelligence in one continuous workflow. While you're there, check out .../build.
​