The Invisible IT Inversion: AI Made Software Faster to Build. The Real Work Moved Behind the Scenes
Posted on July 9, 2026
Originally published: 9 July 2026
Last updated: 14 November 2026
Research for this article commenced around Easter 2026, with an increasing focus on the external factors affecting software development, cybersecurity, digital infrastructure and business operations. Originally published on 9 July 2026, the analysis has been progressively expanded following significant developments. Substantive updates are dated throughout, with a complete revision history at the end.
A business website can look exactly the same today as it did six months ago. The customer sees the same pages, the same features and the same familiar experience. From their perspective, very little has changed.
Behind the scenes, however, the technology supporting that website may have undergone dozens of important changes. Security vulnerabilities discovered and patched. Software dependencies updated. APIs revised. Authentication systems strengthened. Hosting infrastructure maintained, monitored and protected against threats that didn't exist when the website was first built. And increasingly, all of this is happening while artificial intelligence makes it possible to develop new software faster than ever before.
This is what I call the Invisible IT Inversion.
AI is dramatically accelerating software creation, but it hasn't eliminated the responsibility of securing, maintaining and operating what we build. Instead, the rapid adoption of AI-assisted development is shifting more of the operational pressure towards verification, security, integration management and continuous maintenance.
For business owners and executives, this creates an important distinction between the cost of building technology and the cost of keeping that technology reliable, secure and commercially useful. For digital agencies, developers and technology service providers, it introduces another challenge: how to adapt their operating models, allocate resources and maintain commercially sustainable relationships with clients whose expectations are also changing.
Software is becoming easier to create. That doesn't necessarily make it easier or cheaper to own.
Where This Research Started: The Changing Nature of IT Work
Around Easter 2026, I began researching what appeared to be a significant shift in the relationship between software development, ongoing maintenance and operational risk. Initially, the focus was on how AI-assisted coding tools were changing the development process. The productivity improvements were becoming increasingly apparent, particularly in generating application components, writing functions, developing integrations and reducing repetitive programming work. But the more interesting question was what happened after that code was generated.
As the research developed, I began expanding the scope beyond software production itself, increasingly monitoring external factors that could materially affect the cost, complexity and reliability of operating digital systems. These included vulnerability disclosure rates, the speed of cyber exploitation, international cybercrime, telecommunications and communications-platform risks, Australian regulatory developments, workforce capability and broader economic conditions.
By the time the original article was published in July, the underlying tension was already evident. Software production was becoming faster and more accessible, while the environment in which that software needed to operate was becoming increasingly demanding, interconnected and exposed to disruption. What interested me wasn't simply that cybersecurity threats were increasing or that AI was changing development. Both were already well-established trends.
The more important question was how these forces were changing where the engineering work occurred, who carried the responsibility and what businesses would ultimately pay to maintain reliable digital operations. The subsequent developments documented in this article have provided further evidence of that shift, while also revealing opportunities that I believe are just as significant as the risks.

The AI Development Paradox: Faster Production Creates New Bottlenecks
For decades, a major constraint in software development was the time required to create the software itself. Developers translated requirements into code, designed interfaces, connected databases, built integrations and worked through testing and deployment. The time required to produce functioning software was often a substantial component of project cost.
AI-assisted development has changed that equation. Developers can now generate application components, produce working code examples, create tests, document functionality and explore different technical approaches in considerably less time. For straightforward applications and well-defined tasks, these improvements can be substantial. However, generating code is only one part of delivering dependable software.
AI-generated code still needs to be reviewed for correctness, security, compatibility, maintainability and suitability for its intended application. A technically plausible implementation can introduce a security weakness, misunderstand business rules or break a dependency elsewhere in the system. The faster code can be produced, the more important these downstream activities become.
This creates a potential bottleneck in the development pipeline. Developers can generate changes more quickly than an organisation can safely review, test, integrate and deploy them. It's particularly relevant where AI is being used to develop custom integrations, plugins, automation workflows or business applications. These systems often interact with existing software, databases and external services that the AI tool may not fully understand.
For example, a headless WordPress application using GraphQL or custom API integrations may have multiple layers of functionality, authentication and data handling. As I've previously explored in my article on Gato GraphQL and headless WordPress, modern architectures can create powerful opportunities for connecting applications and automating workflows. However, those integrations also require ongoing consideration of permissions, data consistency, compatibility and security.
The practical distinction is between code produced and working software successfully delivered. A development team that generates twice as much code hasn't necessarily doubled its productivity if the additional output requires extensive rework, introduces defects or increases the ongoing maintenance burden.
AI can create genuine operational leverage. But measuring that leverage requires looking beyond the speed of the initial development task and examining the entire software lifecycle. This also changes how businesses should evaluate development costs. A faster initial build may reduce the upfront investment, but that saving can be diminished if poor architecture, excessive integrations or inadequate testing create ongoing operational problems.
Software Vulnerabilities Are Changing the Maintenance Equation
The second major pressure comes from the volume and speed of software vulnerabilities. WordPress provides a useful example because it supports everything from relatively simple business websites to complex ecommerce platforms, membership applications and customised business systems.
According to Patchstack's State of WordPress Security in 2026, published in February and examining 2025 data, researchers identified 11,334 new vulnerabilities across the WordPress ecosystem. That represented a 42% increase compared with 2024, with approximately 91% of the identified vulnerabilities affecting plugins. On average, that's around 218 newly reported vulnerabilities every week across the ecosystem.
It doesn't mean every WordPress website is exposed to hundreds of new vulnerabilities each week. Exposure depends on the components installed, their versions, configurations and the relevant security weaknesses. Nevertheless, the numbers demonstrate the scale of security information that developers, vendors, hosting providers and website operators must continually process.
The more significant finding concerns the time available to respond. Patchstack reported that 46% of vulnerabilities had not received a developer fix before public disclosure. Its research also found a weighted median time of approximately five hours between disclosure and observed exploitation among a subset of heavily exploited vulnerabilities. That five-hour figure does not describe every vulnerability; it reflects particularly attractive targets experiencing substantial exploitation activity. However, it demonstrates how quickly serious threats can develop once technical information becomes publicly available.
Historically, many website maintenance arrangements centred on monthly updates, periodic backups and reactive troubleshooting. Those activities remain useful, but they are not sufficient on their own for business-critical systems facing actively exploited vulnerabilities. A serious vulnerability might emerge shortly after a scheduled maintenance window. There may be no developer patch available, or an immediate update could create compatibility problems with other components.
In these circumstances, maintaining the system may involve temporary security controls, restricting affected functionality, implementing application-level protection, testing updates in staging environments and coordinating an emergency deployment. Modern maintenance increasingly requires continuous awareness of risk, rather than occasional attention to software versions.
The distinction matters commercially because monitoring, assessment, emergency response and controlled remediation are different activities from routine updates. They require appropriate systems, processes, expertise and accountability, even when no visible feature has changed.
Cybersecurity Is No Longer Confined to the Website or Application
As digital systems become more interconnected, the potential attack surface extends far beyond the application itself. A business may operate a secure website but still be exposed through compromised employee credentials, weak authentication, external integrations, email infrastructure, communications providers or third-party access. AI-assisted social engineering adds another dimension.
A study published in Expert Systems with Applications in June 2026 evaluated the effectiveness of automated AI-generated spear-phishing against 101 human participants. In the controlled experiment, fully AI-generated phishing emails achieved a 54% click-through rate, matching the human-expert group and substantially exceeding the 12% control group receiving generic phishing messages.
These results should not be mistaken for the average success rate of phishing campaigns in the real world. They do, however, illustrate how effectively AI can automate personalised deception. An attacker no longer needs to manually research every target or spend considerable time constructing a convincing message. AI can assist in gathering information, tailoring language and generating communications designed to appear relevant and credible.
Importantly, the same research also demonstrated the potential for AI-assisted phishing detection, illustrating that these technologies create opportunities for defensive improvement as well as offensive misuse. For businesses, the implications extend into identity verification, staff training, payment authorisation, account recovery and the controls surrounding sensitive information.
The problem becomes particularly serious when fraudulent communications appear to originate from legitimate businesses or trusted service providers. Protecting an application therefore means examining not only the software itself, but also how people and other systems are authorised to interact with it.
The Invisible Infrastructure Supporting Everyday Business
One of the most overlooked aspects of modern digital operations is the amount of infrastructure operating behind otherwise simple customer interactions. Consider a customer submitting an enquiry through a website. The visible process might take less than a minute, yet the information could pass through an application server, database, CRM platform, automation service, transactional email provider and SMS notification gateway.

Each component introduces dependencies, credentials, permissions, configuration requirements and potential points of failure. Communications Platform as a Service (CPaaS) providers such as Twilio, alongside transactional email services including SendGrid and Mailgun, have become important components in this environment. These platforms support functions such as automated notifications, one-time passwords, password resets, account verification and operational alerts.
When a connected service is compromised, misconfigured or unavailable, the resulting disruption can affect numerous businesses relying on that infrastructure. This dependency risk isn't theoretical. In August 2022, attackers compromised internal Twilio systems through employee phishing. The incident ultimately affected 163 customer organisations, as documented in subsequent reporting on Twilio's security investigation. It demonstrated how an intrusion involving a shared technology provider can create consequences across multiple downstream businesses.
More recently, Twilio's March 2026 Fraud Update documented increasingly sophisticated abuse of legitimate communications channels. The company reported that phishing volume involving certain marketing email templates had nearly doubled during the first quarter of 2026. This was a reported increase in a particular phishing method, not evidence that Twilio experienced another equivalent breach.
Another relevant development occurred on 1 July 2026, when Australia's mandatory SMS Sender ID Register commenced. Under the rules, organisations using branded alphanumeric SMS sender IDs must register those names through participating telecommunications or messaging providers. Messages using unregistered sender IDs can have the displayed sender replaced with Unverified.
For Australian digital agencies managing CRM platforms, marketing automation and client communications, the requirements introduce practical compliance and integration responsibilities. The rules apply to branded alphanumeric sender IDs rather than ordinary messages sent from mobile numbers. The website might still function. The CRM might still work. The API might still connect. But the communications process may no longer operate in the way the business intended.
Maintaining modern software also means maintaining the external relationships, permissions, standards and infrastructure that allow it to function properly.
RESEARCH UPDATE · 10 AUGUST 2026
WordPress Security Reinforces the Maintenance Problem
The weeks following the original July publication provided further examples of how quickly software security responsibilities can change. On 17 July 2026, WordPress released version 7.0.2 to address one critical and one high-severity security issue. The vulnerabilities included a REST API issue involving batch-route handling and SQL injection that could lead to remote code execution under relevant conditions.
The WordPress security announcement recommended immediate updates, with automatic updates enabled for affected installations because of the severity. Less than three weeks later, on 6 August 2026, WordPress released version 7.0.3 with additional security fixes, again recommending immediate updates.
These releases illustrated the practical reality facing website operators. Even widely used and actively maintained platforms can require urgent security action outside normal maintenance schedules. A second development reinforced the wider infrastructure risk.
On 16 July 2026, the Office of the Australian Information Commissioner published findings concerning the 2025 Qantas data incident, which affected approximately five million Australians following a social engineering attack involving an overseas third-party provider. The regulator did not find sufficient evidence of a likely Privacy Act contravention to justify further investigation. However, the report provided a useful examination of third-party risk, security controls, incident response and the limitations of preventative measures.
These examples reinforced two important conclusions. The first is that security maintenance cannot rely exclusively on predictable release schedules. The second is that digital resilience depends on both technical controls and the wider operational arrangements surrounding a system.
For agencies and service providers, that means understanding which responsibilities belong to the hosting provider, application developer, software vendor, communications provider and client, and ensuring those responsibilities are clearly documented. It also creates an opportunity to improve service delivery: automated vulnerability monitoring, staged testing and well-defined emergency procedures can help smaller teams manage larger technology portfolios without relying entirely on manual intervention.
The question is no longer whether agencies should provide ongoing technical maintenance. It's whether their operating arrangements are capable of responding appropriately when routine maintenance becomes an urgent security responsibility.
RESEARCH UPDATE · 30 SEPTEMBER 2026
Australia's Economic Conditions and Agency Pressures
By September, the research was increasingly pointing towards a broader commercial challenge. Businesses were facing pressure to improve productivity, control operating costs and adopt emerging technologies, while digital service providers were dealing with increasingly complex infrastructure and maintenance obligations.
Australian Industry Group analysis of ABS data found that Australia's whole-economy labour productivity declined by 0.6% during the 2024–25 financial year. At the same time, recruiting and retaining appropriate staff remained costly.
On 15 September 2026, Employment Hero published research indicating that while 67% of Australian workers would consider a suitable opportunity, only 6% were actively applying for jobs. Its survey-based estimates suggested that unfilled roles were contributing to average annual revenue losses of approximately $160,000 per business. That figure is an estimated average across surveyed businesses, rather than an established loss per individual vacancy or a specific measure of technology recruitment.
For smaller digital agencies, the challenge is particularly relevant. Building a large team of developers, infrastructure specialists, security professionals, account managers and technical support staff can create substantial fixed costs. At the same time, clients are increasingly aware that AI can reduce the time required to produce websites, applications, marketing content and other digital deliverables.
This creates a difficult commercial situation when service providers are simultaneously facing pressure to lower development prices while increasing their investment in security, maintenance and ongoing operations.
Australian Telecommunications and Identity Risks
On 3 September 2026, the Australian Communications and Media Authority announced that Telstra had paid a $277,200 penalty after an investigation identified failures involving required customer identity-authentication processes and protections against mobile number fraud. The findings included unauthorised SIM swaps and inadequate safeguards for customers at risk of fraud.
The incident followed years of heightened attention to telecommunications security, including the substantial consequences of the September 2022 Optus data breach and separate Telstra third-party data exposures. These were different incidents involving different circumstances, but they reinforced the importance of maintaining effective identity controls, staff procedures and operational accountability. They also illustrated why the costs associated with digital security extend beyond software licences and technical infrastructure.
The Changing Economics of Digital Agencies
The disruption was becoming visible in the advertising and marketing industry as well. On 9 September 2026, M+C Saatchi announced that a proposed management buyout of its Australian and New Zealand advertising business would not proceed. The affected advertising operations faced an uncertain future, with the company entering discussions about client transitions to other parts of the wider group. Its separate government-services business was unaffected.
The circumstances involved commercial performance, ownership decisions and an unsuccessful transaction. It would be inaccurate to attribute the difficulties solely to AI. Nevertheless, the case demonstrated the consequences of operating in an industry where traditional business models are under pressure and commercial flexibility has become increasingly important.
A wider shift was also evident in the use of AI within agencies. In June, Forrester reported that approximately nine in ten US marketing agencies were using generative AI, with around half using agentic AI for marketing execution. However, its research also warned that focusing exclusively on efficiency and cost reduction could undermine creativity, differentiation and long-term marketing effectiveness.
That distinction matters. AI can reduce the cost and time required for particular activities. But producing more content, completing more tasks or automating more workflows doesn't automatically create better client outcomes. The opportunity lies in combining automation with technical expertise, commercial judgement and a service model that can demonstrate real value.
RESEARCH UPDATE · 31 OCTOBER 2026
A Halloween Security Scare — But It's Not the End of the Web
October delivered one of the most significant examples of the infrastructure and maintenance risks examined throughout this research. Given the timing, the latest vulnerability disclosures arrived with all the ingredients of a Halloween cybersecurity horror story: critical software flaws, remote code execution, compressed response windows and concerns about the security of a technology underpinning a substantial portion of the internet.
But before we start digging graves for every PHP-powered website, it's worth examining what actually happened.
PHP2Shell: The Vulnerabilities Beneath the Application
On 21 October 2026, the PHP project released emergency security updates addressing two critical vulnerabilities, CVE-2026-109618 and CVE-2026-109619, collectively associated with the name PHP2Shell. Both vulnerabilities received a CVSS v4.0 severity score of 9.2 and could, under relevant conditions, allow unauthenticated attackers to execute code on affected servers. Unlike vulnerabilities confined to individual website plugins, these weaknesses affected the underlying PHP runtime used by a substantial proportion of the web.
The PHP Foundation's official announcement confirmed the release of patched versions PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. The Foundation described the extensive investigation, technical coordination and difficult compatibility decisions involved in addressing the vulnerabilities.
These were serious security issues, but the distinction between a widely deployed technology and a universally exploitable vulnerability is important. A website using PHP was not automatically compromised simply because the vulnerabilities existed. Actual exposure depended on the runtime version, deployment configuration, relevant attack conditions and available mitigations.

That doesn't reduce the urgency of remediation; it provides the context needed to respond professionally rather than react to alarming headlines.
AI Was Also Part of the Solution
One of the more encouraging aspects of the PHP2Shell disclosure was the role of AI-assisted security research. The PHP Foundation documented how researchers and engineers used increasingly sophisticated tools to investigate vulnerabilities, validate findings and support the development of security fixes. This is particularly important because AI was contributing to the defensive research effort, not simply creating additional threats.
The same technologies accelerating software development were also helping researchers identify weaknesses, validate technical findings and improve security processes. However, the Foundation's account demonstrated that advanced tools cannot eliminate the need for experienced engineers to understand complex codebases, evaluate compatibility risks and make difficult implementation decisions.
The broader operational implications were substantial. Businesses and infrastructure providers needed to identify affected environments, assess their exposure, coordinate updates, test compatibility and ensure that critical applications continued operating correctly. For website owners, the incident highlighted a particularly difficult aspect of digital maintenance: serious vulnerabilities can emerge in foundational components that are largely invisible to end users.
A website can be fully functional, visually unchanged and apparently operating normally while the underlying infrastructure requires immediate security attention. PHP2Shell provided a clear example of the Invisible IT Inversion: the visible application remained the same, but the engineering work required to keep it secure suddenly became critical.
The Positive Side of the Scare
The incident also demonstrated how modern cybersecurity capabilities are improving. Security researchers, software maintainers, infrastructure providers and specialist vendors were able to coordinate investigations, develop fixes and help organisations respond to a technically complex problem. AI-assisted research, automated testing, vulnerability intelligence and coordinated security processes all contributed to the response.
For digital agencies and managed service providers, these developments present opportunities to strengthen their own operational capabilities. Automated dependency inventories, security monitoring, staging environments and controlled deployment pipelines can improve both response times and service consistency. Temporary virtual patching may also help with some vulnerability classes, although the PHP Foundation made clear that updating PHP was necessary to fully address these particular flaws.
These systems won't eliminate cybersecurity risks or remove the need for experienced professionals. They can, however, reduce repetitive work and help technical teams manage increasingly complex environments more effectively. So while PHP2Shell provided plenty of Halloween headlines, the longer-term story isn't necessarily one of inevitable disaster.
The threat environment is evolving, but so are the tools and expertise available to defend against it. The opportunity is to use those capabilities intelligently, establish appropriate responsibilities and make security management part of normal digital operations rather than something addressed only when a crisis emerges.
RESEARCH UPDATE · 14 NOVEMBER 2026
Cyber Escalation, AI Engineering and Industry Restructuring
The first two weeks of November brought further developments across cybersecurity, software engineering and the commercial technology sector. Taken individually, these events involved very different organisations and circumstances. Viewed together, they provided additional evidence that technology production, operational security and business economics were changing at the same time.
North Korea's Binance Attack: The Scale of Modern Cyber Risk
On 29 October 2026, cryptocurrency exchange Binance suffered a major cyberattack involving the theft of more than US$6 billion in virtual assets. The US Federal Bureau of Investigation attributed the attack to North Korea's TraderTraitor cyber operation.
On 12 November 2026, the US Department of Justice announced coordinated action against North Korean cryptocurrency theft operations. Government and private-sector efforts had frozen more than US$2.7 billion of the stolen Binance assets, while court authorisation had been obtained to seize or restrain more than US$1.2 billion of that amount. These figures represent different stages of the recovery and enforcement process and should not be treated as interchangeable amounts.
The announcement also described a separate North Korean social engineering campaign involving fraudulent video meetings, impersonation and malware designed to compromise trusted access. Although the scale of these operations is far removed from a typical Australian business application, the broader security lessons are relevant.
Sophisticated social engineering, malicious software and the compromise of trusted access can undermine otherwise substantial technical infrastructure. For businesses, this reinforces the importance of managing identity, credentials, external integrations and authorised access, rather than treating cybersecurity exclusively as a software patching problem.
AI Engineering Research: The Verification Bottleneck Becomes Measurable
Additional engineering research has provided stronger quantitative evidence of the development pressures discussed in the original July analysis. LinearB's 2026 engineering benchmarks reported that agentic AI-generated pull requests waited approximately 5.3 times longer for initial review than unassisted pull requests at the 75th percentile. The same research reported that approximately 32.7% of AI-generated pull requests were accepted, compared with 84.4% of unassisted requests.
Separately, Faros AI's 2026 engineering analysis, drawing on telemetry from approximately 22,000 developers across 4,000 teams, identified increasing downstream pressure in development workflows. Its comparisons between high and low AI-adoption environments reported substantially longer review times, increased code churn and greater operational friction, despite improvements in initial development output. One particularly striking finding was a 441.5% increase in median time spent in pull-request review in the study's comparison of higher and lower AI adoption.
These are observational findings from particular engineering datasets. They do not establish that AI alone caused the differences or that every organisation will experience the same outcomes.

Nevertheless, they reinforce an important part of the original thesis: accelerating code production creates limited commercial value if the organisation cannot safely review, integrate and deploy the additional output.
The emerging challenge is not simply helping developers write software faster. It is improving the efficiency and reliability of the complete development and delivery process.
Canva's November Restructuring: Even Established Technology Leaders Are Adapting
In early November, Australian design software company Canva announced a significant organisational restructure affecting approximately 1,270 employees. The changes included redundancies, changes to existing roles and requests for some employees to relocate. The number affected should not be confused with the number ultimately made redundant.
Canva's co-founder and CEO, Melanie Perkins, explained that the company's growth had slowed for the first time in a decade and that its existing organisational structure was no longer appropriate for the next phase of its development. The restructuring involves consolidating Product, Growth and Platform functions, reducing management layers, broadening responsibilities and returning to smaller teams with greater autonomy.
However, the more interesting question is what these changes reveal about the technology market in which Canva operates. For more than a decade, Canva's core proposition was making graphic design and visual content creation more accessible. It helped businesses and individuals produce professional-looking designs without requiring the same level of specialist software knowledge or technical capability traditionally associated with graphic design.
Now, AI is taking that accessibility considerably further. Many activities that previously required selecting templates, manually arranging visual elements, editing images and developing marketing materials can increasingly be accomplished through natural-language instructions. A business owner can describe what they want and use AI tools to generate presentations, promotional graphics, social media content, documents and other creative assets directly.
These capabilities aren't limited to Canva. They are emerging across AI platforms, specialised creative tools and increasingly capable custom-built agents. For many routine production tasks, the result can be significantly faster and less expensive than traditional manual workflows. The more important competitive issue is that AI is changing the value of the underlying activity itself.
Canva recognised this shift earlier in 2026 when it introduced Canva AI 2.0, expanding its platform towards conversational design, agentic workflows and AI-assisted content production. The company is therefore participating in the transition, not simply being displaced by it.

It would be inaccurate to conclude that AI alone caused Canva's November restructuring. Its stated reasons included slowing growth, organisational complexity and changing technological requirements. Nevertheless, the development illustrates a wider commercial reality: if AI makes activities that once required specialised platforms or professional services available directly to users, established providers must reconsider where they create value.
That applies to graphic design, digital marketing, website development, content production, business automation and numerous other technology-enabled services. Canva's restructuring doesn't establish that its business model is obsolete. It demonstrates that even highly successful technology companies must adapt when the capabilities available to customers and the economics of delivering those capabilities change.
For digital agencies, the lesson is particularly relevant. The question is no longer simply whether AI can perform an existing task. It's what expertise, service and commercial value remain important when that task becomes faster, cheaper and more accessible.
The Opportunity for Digital Agencies: Getting Ahead of Client Expectations
One of the most important commercial implications of this research concerns the relationship between digital agencies and their clients. As AI makes software development, content production and digital marketing activities faster and more accessible, client expectations will inevitably change.
Business owners are already being exposed to increasingly capable tools that promise faster delivery, lower production costs and automation of tasks that previously required specialist providers. At the same time, agencies are carrying increasing responsibility for cybersecurity, infrastructure maintenance, integrations, system reliability and ongoing technical support.
If these changes aren't communicated effectively, a gap can emerge between what clients believe they're paying for and the work their service provider is actually undertaking. This creates a commercial risk, but it also presents an opportunity.
Client Communication Is Becoming a Competitive Advantage
My experience across business operations, technology, CRM systems and customer relationship management has reinforced the importance of understanding what clients are actually trying to achieve, documenting their requirements and maintaining clear communication throughout service delivery. That principle becomes particularly important when the nature of the service itself is changing.
A client who sees no visible changes to their website may reasonably question the value of ongoing maintenance expenditure. They may not understand the security monitoring, dependency management, infrastructure updates or emergency remediation activities being performed behind the scenes. The solution isn't to overwhelm clients with technical explanations or create additional complexity to justify the work; it's to communicate the responsibilities and outcomes that matter to them.
For most business owners, these outcomes are straightforward: their systems remain available, their information is protected, their customers can transact and communicate reliably, and the technology continues supporting their business objectives. Digital agencies have an opportunity to take the initiative by reviewing their client portfolios, explaining how service requirements are changing and aligning their ongoing responsibilities with each client's actual risk and operational needs.
That might involve reviewing maintenance arrangements, identifying previously unmanaged dependencies, clarifying security responsibilities or establishing appropriate response procedures for critical incidents. The commercial conversation should focus on what is genuinely required, what can be improved through automation and what additional responsibilities need to be resourced.
Clients should understand what is changing, why it matters, who is responsible and what value they're receiving.
Resource Allocation, Pricing and Service Expectations
A second opportunity concerns how agencies allocate their own resources. Traditional agency models often separate development, design, marketing, account management, technical support and infrastructure responsibilities into distinct activities or departments. That approach can be appropriate for larger operations, but it can also introduce unnecessary handovers, duplicated work and fragmented accountability.
AI-assisted workflows and improved automation create opportunities to reconsider how those responsibilities are organised. For example, routine vulnerability monitoring, dependency checks, test generation, documentation and operational reporting can increasingly be supported by automated systems. That doesn't eliminate the need for experienced professionals; it allows those professionals to concentrate on architectural decisions, complex problem-solving, security judgement, client relationships and higher-value activities.
The same principle applies to client account management. Rather than responding only when something goes wrong, agencies can use better monitoring and reporting to identify emerging issues, communicate risks and recommend improvements before they become expensive problems.
This is closely connected to the principles I've previously discussed around making task management systems genuinely useful. Technology should make responsibilities, priorities and next actions clearer, rather than creating more administration for everyone involved.
However, commercial responsibility works both ways. If automation substantially reduces the effort required to deliver an existing service, those efficiency improvements should be considered when reviewing internal costs and client pricing. Equally, if the agency is assuming additional security, monitoring or operational responsibilities that were never included in the original agreement, those obligations should be properly defined, resourced and commercially agreed.
The objective is not to justify higher fees simply because technology is becoming more complicated. It's to develop a service model where responsibilities, costs and expected outcomes remain aligned.
AI Is Also Creating the Tools Needed to Solve the Problem
Although much of the research has focused on emerging risks and operational pressures, the opportunities developing alongside them are substantial. AI is not simply accelerating the creation of software and increasing the volume of code requiring verification. It is also becoming an increasingly useful part of the infrastructure used to secure, maintain and operate that software.
Smarter Vulnerability Monitoring and Remediation
AI-assisted tools can help developers and security teams identify vulnerable components, analyse code, investigate unusual behaviour, review security reports and prioritise issues requiring attention. Combined with vulnerability intelligence, automated dependency monitoring, web application firewalls and controlled deployment systems, these capabilities can reduce the time spent identifying and responding to routine problems.
Some platforms already provide automated protection against known exploit patterns while permanent software fixes are being prepared. The broader security ecosystem is also becoming more coordinated, with software maintainers, hosting providers and security researchers using increasingly sophisticated systems to identify threats and distribute mitigations.
These developments won't eliminate the need for emergency maintenance or experienced security professionals. They do, however, create opportunities to manage larger and more complex environments with greater consistency and efficiency.
AI-Assisted Infrastructure and Software Maintenance
AI can also assist with routine operational work that has historically consumed substantial developer and system administrator time. Examples include reviewing logs, generating test cases, documenting infrastructure changes, identifying dependency conflicts, checking configuration differences and helping engineers investigate unexpected application behaviour.
More advanced workflows can combine monitoring systems, development repositories, testing environments and deployment tools to support the investigation and controlled remediation of issues. The distinction between assisting a technical professional and autonomously modifying production infrastructure remains important.
Not every automated recommendation should result in an automatic change. Appropriate permissions, testing, approval processes, audit trails and rollback arrangements remain essential, particularly for business-critical applications. But when implemented properly, these systems can reduce repetitive labour and help experienced teams respond more effectively.
This reflects the broader principles behind continuous improvement: identify the problem, test appropriate changes, assess the results and continue improving the process. AI creates opportunities to accelerate that cycle, provided the underlying systems remain controlled and the results are properly evaluated.
Custom AI Agents and the Changing Economics of Specialist Digital Services
Another area of opportunity is the development of purpose-built AI agents capable of supporting increasingly specialised business functions. Traditional digital marketing and technology services often require separate tools, subscriptions, staff and manual processes for research, content development, reporting, campaign administration, data analysis, workflow management and customer communications.
Customised AI agents can increasingly bring several of these activities into coordinated workflows, using approved data sources, defined instructions, integration capabilities and human oversight. For suitable tasks, this can improve turnaround times, reduce repetitive work and make specialist capabilities more accessible to smaller organisations.
But cost, speed and quality should be evaluated independently. Producing content more quickly doesn't guarantee better marketing performance. Generating a report automatically doesn't guarantee accurate conclusions. And deploying an AI agent doesn't necessarily remove the need for strategic expertise or professional review.
The commercial opportunity lies in replacing unnecessary manual activity with more efficient systems while preserving the judgement and accountability needed to deliver useful outcomes. For digital agencies, this may support a shift away from selling labour-intensive production tasks towards delivering measurable business outcomes, managed infrastructure and specialised technology-enabled services.
For clients, it may mean accessing capabilities that were previously too expensive or operationally difficult to justify. The businesses that benefit most will be those that understand which activities can safely be automated, where human expertise adds genuine value and how the resulting improvements should be reflected in their service models.
A More Sustainable Operating Model for the Next Generation of Digital Agencies
The convergence of AI development, cybersecurity, infrastructure complexity and changing commercial expectations doesn't mean every agency must rebuild its operations from scratch. It does suggest that the traditional separation between building something and maintaining it deserves reconsideration.
A more sustainable approach begins with visibility. Agencies need to understand the systems they manage, the dependencies those systems rely upon and the responsibilities they have accepted. From there, they can introduce appropriate monitoring, automation and risk-based maintenance arrangements while retaining experienced oversight of significant technical decisions.
Client communication should become part of that operating model rather than an activity that occurs only during sales, renewals or disputes. Regular, useful reporting can explain what has been maintained, which risks have been addressed, what improvements have been implemented and whether the service continues to meet the client's requirements.
That doesn't mean producing lengthy technical reports that clients are unlikely to read. It means making important information understandable, demonstrating accountability and helping clients make informed decisions.
Agencies also need to examine how their own costs are changing. AI subscriptions, infrastructure, API usage, security platforms, technical staff, testing and support all form part of the operating environment. Where automation reduces manual work, the resulting capacity can be redirected towards higher-value services, improving quality or strengthening business resilience.
Where new responsibilities create genuine additional costs, those should be identified rather than absorbed indefinitely through poorly defined service arrangements. The goal should be a simpler, more transparent and more resilient service model.
The Practical Priorities
For agencies considering how to respond, I would prioritise six areas:
- Understand and document the technology environment. Maintain a reliable inventory of websites, applications, infrastructure, plugins, integrations and external services. Identify dependencies, system owners and the responsibilities attached to each component.
- Introduce continuous monitoring and risk-based maintenance. Use appropriate automation to identify security vulnerabilities, dependency changes and infrastructure issues. Distinguish routine maintenance from critical vulnerabilities requiring urgent intervention.
- Improve testing, deployment and recovery processes. Use staging environments, automated tests, backups, rollback arrangements and documented procedures to reduce the risk of introducing new problems while resolving existing ones.
- Review client agreements and service expectations. Clearly distinguish development, routine maintenance, security monitoring, emergency response and additional technical responsibilities. Align service commitments with the actual resources and capabilities available.
- Use AI to improve capacity rather than simply increase output. Prioritise tasks where automation can reduce repetitive work, improve consistency and assist experienced professionals. Measure the cost of reviewing and correcting automated output as well as the time saved generating it.
- Strengthen proactive communication and account management. Explain relevant risks and improvements in plain language, establish clear responsibilities and review whether the service continues to deliver the outcomes the client requires.

These priorities are not a universal implementation checklist. The appropriate level of investment will differ substantially between a straightforward informational website and a complex ecommerce, financial or business-critical application. The objective is to match operational controls to genuine risks and requirements, avoiding unnecessary complexity while maintaining appropriate professional standards.
Good systems reduce uncertainty. Good automation reduces unnecessary work. Good client relationships ensure that both translate into outcomes people actually value.
Conclusion: The Inversion Is Creating a Different Kind of Opportunity
The developments throughout 2026 have reinforced the central observation behind this research. AI is making software faster and more accessible to create, while cybersecurity, interconnected infrastructure, regulatory requirements and changing commercial expectations are placing greater importance on what happens after that software is built.
The engineering work hasn't disappeared. Much of it has shifted towards verification, security, integration management and continuous operation. But this isn't simply a story about increasing costs, more vulnerabilities or an industry struggling to keep pace. It's also the beginning of a significant opportunity.
The same technologies accelerating software production are helping organisations identify vulnerabilities, strengthen security, automate maintenance and manage increasingly complex systems with fewer repetitive manual processes. Purpose-built AI agents are opening access to specialised capabilities that previously required substantial resources, while more sophisticated infrastructure and security services are improving what smaller businesses and technology teams can realistically manage.
At the same time, the changing economics of digital services are creating opportunities for agencies willing to reconsider how they allocate resources, structure their services and communicate with clients. There will undoubtedly be further disruption. Established providers will continue adapting, new competitors will emerge and some traditional business models may become increasingly difficult to sustain.
However, the businesses best positioned for the next phase won't necessarily be those producing the most software or adopting the greatest number of AI tools. They will be the organisations that understand their systems, manage their dependencies, automate intelligently, maintain strong client relationships and preserve the professional judgement required to deliver dependable outcomes.
AI is changing how quickly we can build the future. The opportunity now is to use that same capability to make what we build more secure, reliable, efficient and commercially sustainable.
What's Next: A Practical Guide for Digital Agencies — January 2027
Throughout 2026, alongside this research, I've been documenting, developing and testing tools, systems and operational processes intended to help digital agencies navigate the changing environment described in this article. The work has increasingly focused on the practical challenges agencies face when AI-assisted development, cybersecurity responsibilities, infrastructure maintenance, client expectations and commercial pressures converge.
I'm now developing this material into a dedicated Digital Agency Guide, which I aim to publish in January 2027. The guide will move beyond the research and analysis presented here to focus on practical implementation, including AI-assisted development and maintenance workflows, vulnerability monitoring, infrastructure management, automation, resource allocation, cost control, client communication and account management.
It will also examine how agencies can adapt existing service arrangements and introduce new capabilities without creating unnecessary complexity or losing sight of the client's actual objectives. This article will continue to be monitored and updated as significant new research, technical developments, industry changes and relevant events become available. Where new evidence materially changes the findings, those developments will be incorporated and clearly dated.
Register your interest in the January 2027 Digital Agency Guide using the form below to receive notification when it becomes available.
I'm also interested in receiving relevant research, practical experiences, implementation examples and solutions from agency owners, technology professionals and business leaders. These contributions are welcome privately through the contact and social channels linked on this website, or through the comments section below. Substantiated findings that challenge or refine the analysis are equally valuable.
The objective is to continue building a practical, evidence-based understanding of how the digital services industry is changing—and, more importantly, how businesses can adapt successfully.
— Darryn Van Hout
Editorial Record
Article Revision History
Originally published 9 July 2026. Last updated 14 November 2026. Select an update below to read the details.
Major Research, Industry and Commercial Update
Expanded the analysis to include the North Korean cyberattack against Binance on 29 October and subsequent US government action announced on 12 November, additional engineering research concerning AI-generated code review, and Canva's early November organisational restructuring affecting approximately 1,270 employees.
Strengthened the commercial analysis of digital agencies, including client communication, resource allocation and changing service expectations. Expanded the opportunities arising from AI-assisted operations and introduced the forthcoming January 2027 Digital Agency Guide.
PHP2Shell and Cybersecurity Update
Added analysis of the PHP2Shell vulnerabilities disclosed on 21 October, including their implications for runtime security, infrastructure dependencies, vulnerability response and patch management.
Introduced the Halloween-themed examination of the security scare, distinguishing serious vulnerabilities from exaggerated claims of universal compromise. Expanded the discussion of AI-assisted security research, coordinated remediation and improving defensive technologies.
Australian Business and Industry Update
Expanded the Australian economic and commercial analysis to incorporate September labour-market research, productivity findings, ACMA's Telstra enforcement action and developments affecting the Australian advertising-agency industry.
Further examined the implications for agency operating costs, workforce and resource allocation, service delivery, digital production and changing client expectations.
WordPress Security and Infrastructure Update
Incorporated security developments following the original publication, including the urgent WordPress 7.0.2 release on 17 July and the subsequent 7.0.3 security release on 6 August.
Expanded the discussion of third-party infrastructure risk and the Australian Privacy Commissioner's July findings concerning the Qantas data incident. Considered implications for proactive monitoring, emergency maintenance and service-provider responsibilities.
Original Publication
Published the original Invisible IT Inversion analysis following research commenced around Easter 2026.
Examined the emerging relationship between AI-assisted software development, cybersecurity, vulnerability management, connected digital infrastructure and the changing responsibilities associated with ongoing software maintenance.
- S Prev
- s


