Faster AI coding does not mean faster agency delivery

Agency capacity depends on completing useful client work

A digital agency can produce more code and still struggle to deliver more finished client work. Each additional implementation needs review, approval and communication, placing more demands on senior people. Clients add another constraint: they want smaller commitments that deliver clear business value. Agency capacity therefore depends on choosing useful work and carrying it through to approval.

A digital-agency operator with more than 20 years of experience describes these pressures in CMSWire, calling the web’s current pace of change faster than at any time in “the past two decades.” The operator’s work spans content management, commerce and web applications, through shifts involving social media, mobile, cloud, big data and headless systems, which separate content management from the interfaces presenting it. As an agency operator, the writer has a commercial interest in demand for agency expertise and how that work is priced.

The operator traces the broad generative-AI shift to ChatGPT’s arrival in 2022, following earlier participation in the GPT-3 beta. Acceptance into that program led to team testing and a 2021 piece calling the technology a “weapon of mass creation.” The operator says it took a few years for the consequences for coding and the web’s business model to become clear. Those consequences now shape how agencies organize delivery and what clients buy.

Agent-assisted development expands the developer’s responsibilities

To understand that organizational change, start with the constraint agency processes were built around: scarce coding capacity. Detailed planning, Agile and Scrum approaches to organizing development, and strict change controls helped teams avoid spending that capacity on unclear or changing requirements. Enterprise procurement reinforced this arrangement by favoring defined processes and specialized roles. Predictable delivery depended on agreeing what to build and assigning it to people with closely matched expertise.

Specialization could mean committing to one content management system, or CMS, one framework, or either frontend interfaces or backend systems. The operator recalls developers making early career choices that could determine their language, platform or domain for much of their working lives. Clear requirements and controlled processes supported dependable delivery. But the arrangement disadvantaged developers whose strengths fell outside a narrow, lasting specialization.

Coding agents change this division of work by giving individual developers more implementation capacity. In the operator’s experience, developers gained that capacity without corresponding pay increases, while taking on responsibilities associated with managers, team leaders and architects. The prospect of a “10x engineer” captures the ambition for this expanded output. Achieving it requires communication, delegation and coordination across tasks assigned to agents.

These responsibilities can be uncomfortable for developers who enjoy coding and prefer to avoid management. Directing several coding agents requires decisions about task allocation, sequencing and whether the results meet the intended goal. The operator therefore encourages the team to develop as consultants, educators, communicators and leaders. A developer’s technical strength can remain intact while their preferred work becomes a poorer fit for the job.

For professionals prepared to direct the work, the operator reports that agents reduce some technical complexity and risk. Moving between projects has become easier, while the team depends less on external vendors for particular functions and integrations. More work can stay within the team because developers can direct a wider range of implementation. That independence leaves them responsible for deciding what should happen.

Faster implementation increases demand for senior review

The operator reports substantial gains from this expanded capacity, including faster execution, easier project management and clearer accountability. Migrations feel less daunting, and adopting a technology can take “days not months.” User experience and user interface work also place fewer constraints on what the team can deliver. Together, these gains increase the amount of implementation an agency can undertake.

Design implementation offers a concrete measure of the change. The operator recalls an hour of work in Figma requiring up to 20 hours of implementation; the current experience is fewer than five implementation hours for each design hour. For a team discussing an interface with a client, that reduces the effort needed to turn a proposal into usable software. Responding to design changes becomes faster too.

Yet every completed implementation creates work for someone who must assess and approve it. The operator reports a sharp increase in workload across senior roles, including relationship management, as more work requires acceptance decisions, sign-offs and client communication. Implementation can therefore accelerate while approved delivery remains constrained by senior availability. Before accepting another engagement, an agency has to account for the decisions needed to finish it.

Review becomes harder when an agent expands the scope of a change. The operator describes how “a single prompt” can complete a requested task while proposing improvements elsewhere, leading a developer to alter several application areas at once. A reviewer then has to understand how those changes relate and judge their combined effect. Code volume is only part of the workload; the reviewer must also reconstruct the purpose of the changes.

Reconstructing that purpose becomes difficult when parallel agent work arrives without the original prompt or an explanation of how it developed. Earlier review practices suited limited changes arriving in sequence, whereas simultaneous changes require reviewers to establish which were requested, which arose during implementation and which should be accepted. The operator describes the resulting difficulty as exponential and also sees limitations in version-control practices, the methods used to track code changes, under this pattern of work. Code alone may leave important questions about intent unanswered.

The increase in output also changes how agencies select ideas before implementation begins. The operator argues that high development costs previously discouraged teams from pursuing weak ideas, so cheaper implementation requires more deliberate selection. The operator consequently places greater value on imagination and encourages leaders to seek creative input widely. As implementation becomes easier, judging its usefulness takes a larger share of the work.

Earlier working software changes feedback and security work

The ability to produce working software sooner also changes client feedback. The operator recalls presentations ending with a promise to show something tangible in “the next couple of weeks,” followed by a wireframe, a prototype, a mockup and implementation. Clients had to imagine the result through those intermediate stages. Each stage could serve a purpose, but the sequence required time and patience before clients could experience working behavior.

Faster development gives teams more choice about that sequence. The operator says teams can assess whether the full progression is useful or whether to prototype directly in the application’s code, giving clients working software to evaluate earlier. Earlier feedback can then shape implementation while changes are easier to make. The choice depends on whether each design stage adds enough value to justify its effort.

Clients can now create the starting point themselves by using prompts to generate applications, a practice often called vibe coding. They may bring these prototypes to an agency for further development, and the operator emphasizes the care required before shipping them. A convincing demonstration can begin an engagement while still needing professional work on security and readiness for real use. The agency must establish what it will take to turn that demonstration into dependable software.

Security work also extends to threats against the resulting applications. The operator identifies groups of malicious agents as an additional threat, increasing the effort involved in defending software. Agencies therefore have to prepare quickly generated applications for use while accounting for hostile automated activity. Faster production can increase the security work needed before software enters operation.

Explicit specifications help direct agent work

These review and readiness demands make clear development instructions more valuable. The operator recalls documentation-focused colleagues being treated as outliers for advocating detailed specifications, deeper architecture and more formal development practices. Their proposals often received the response, “we‘d love to do these if we only had the time”. With agents doing more implementation, the operator now describes the attitude as “this is the way AI shines when building.”

The practices behind that shift make intended behavior and structure explicit. Specification-driven development organizes work around a written description of what should be built; domain-driven development organizes software around the business concepts it represents; test-driven development uses tests to guide implementation. Along with carefully planned architecture, these approaches give agents clearer instructions and constraints. The operator’s experience is that agents perform better with that structure.

Clearer instructions help direct implementation, while approval still requires people to judge the result. An agency must therefore consider both how it assigns agent work and how much senior time that work will consume. A client-ready delivery requires review, agreement and a commercially justified result that the client can confidently put into operation. The agency’s commitment has to cover that complete path.

Fewer discovery clicks increase the website’s business importance

The question of useful delivery also reaches the websites agencies build, because clients are reconsidering what those properties contribute to the business. Predictions of decline have appeared before: The Daily Mail published “Internet ‘may be just a fad’” in December 2000, and Wired published “The Web Is Dead” in August 2010. The operator also cites an Economist tombstone image bearing 1989–2025, describing its appearance as “last July.” The present concern centers on how discovery and content production are funded.

Content funding becomes vulnerable when chatbot-generated answers change the relationship between publishers and services using their material. The Verge reported on court documents unsealed in September in the New York Times case against Microsoft and OpenAI. Its account described internal Microsoft concern about a “doom loop”: the company’s AI strategy could damage content suppliers’ economic position and, in turn, harm the web and the models depending on that content. The risk is that those suppliers become less able to keep producing material.

Changes in search behavior create a related problem for businesses trying to attract visitors. CMSWire columnist Pierre DeBois’s 2026 column puts searches ending without a click at 68% overall, rising to 83% for queries that return an AI Overview. A business can consequently gain search impressions while attracting fewer organic visits. Marketing and agency teams need to distinguish visibility within a search interface from visits to the website.

Visits arriving through AI have their own commercial characteristics. Adobe Digital Insights’ 2026 report says AI-referred visits produce over 30% more revenue per visit than non-AI traffic, while those visitors stay about 32% longer. It also reports that a quarter of consumers use ChatGPT to inform a purchase. Adobe has a commercial interest in how businesses invest in digital marketing technology, and these findings concern the behavior and value of referred visitors; total revenue also depends on traffic volume.

With fewer informational visits and more competition for attention, the operator describes the priority as “treasuring the click”. Businesses need to make good use of arrivals while recognizing that discovery increasingly happens within other companies’ interfaces. Social platforms and app stores set their own conditions, while a business directly controls its website. That control gives the site a distinct role in serving visitors once they arrive.

The operator sees this role making websites more central to carrying out the business. As external interfaces handle more discovery, the site’s value depends increasingly on what it enables the company and its customers to do. For an agency, choosing what to build therefore requires close attention to the client’s working processes and desired outcomes. Those business requirements also provide a basis for services that can survive changes in AI products.

Agency services need a durable business purpose

A service built around a temporary model limitation faces a different problem. The operator describes a recurring sequence in which a team identifies something a leading AI model cannot do, develops a custom solution and then sees the model vendor add that capability directly. The custom work may meet an immediate need. Its distinct advantage, however, depends on a limitation the vendor is actively removing.

PDF handling gave the operator a concrete example. When ChatGPT could not work with customers’ own PDF data, the operator’s team and other teams built custom solutions. A subsequent press release announced native support, and the workarounds were replaced shortly afterward. The opportunity lasted “only a short time,” showing how quickly a useful technical response could lose its purpose as a separate service.

Waiting for marketing-technology vendors to interpret developments creates another dependency. The operator says agencies that relied on those vendors to explain changes and adjust their offerings fell behind. Both waiting for guidance and building around a temporary capability gap leave an agency’s direction dependent on another company’s product decisions. Agencies need their own understanding because clients need advice while vendors are changing their products.

That understanding can still lead to custom development when a business requirement calls for it. The operator places the more durable value in understanding the requirement, connecting it to a workable process and delivering a sound solution with limited lock-in, meaning dependence on a particular tool or provider. A temporary model limitation may determine how the team builds the solution today. The continuing business requirement gives the client a reason to keep paying for the service.

Clients want smaller commitments and more selective expertise

Turning that business requirement into an engagement starts with questions about time and cost. The operator recalls estimates that once varied within a reasonably predictable range; estimates for the same project can now run from “a few days to a few months.” Clients naturally favor the shorter proposal. Agencies therefore need to explain how their capabilities support a credible commitment.

Explaining that capability also raises a pricing problem because some of its value comes from preparation outside the engagement. Training, research and development can help a team complete work more effectively, while hourly billing ties revenue to time recorded on the project. Value-based pricing, which charges according to the benefit delivered, competes with that established model. The agency has to decide how to capture the value of capability developed before the client’s work began.

Greater implementation confidence can help the agency secure approval for that work. The operator identifies platform migrations and system integrations as areas previously associated with budget overruns, where reduced technical risk now helps teams proceed with deals. This confidence makes the proposal less risky. Pricing and profit still depend on the engagement’s commercial terms.

Clients’ scrutiny of those terms extends to the platforms agencies recommend. The operator sees greater skepticism toward maturity roadmaps for digital experience platforms, or DXPs, alongside closer examination of operating expenses and vendor licensing agreements. Price-conscious buyers need a clear reason to accept additional tools and complexity. A proposed investment must explain the practical benefit well enough to justify the purchase and its continuing cost.

Preparing for those conversations requires agencies to build their own understanding. The operator describes a cultural change combining individual exploration with direct executive participation: executives investigate technologies, contribute ideas and encourage shared learning. Experiments can proceed without strict return-on-investment expectations for each one. That room to learn helps the agency understand changes before making recommendations to clients.

Client investment then requires a different discipline from internal exploration. The agency is developing its ability to make informed decisions, while the client needs a business case for committing money and effort. The operator reports that stronger business-process modeling and automation have revealed previously unrealized project value, helping work receive approval. This has shifted the agency’s positioning toward business solutions led by automation.

As agencies develop that expertise, clients are handling more everyday digital work themselves. The operator says clients increasingly bring agencies in to supplement internal capabilities while asking for more strategic insight and guidance on good practice. Their greater routine independence creates a more selective role for outside specialists. Agencies need to explain where their contribution will change a decision or improve an outcome.

Helping clients secure internal approval becomes part of that contribution. The operator reports that projects require stronger business cases and that clients often need help conducting the conversations necessary to obtain approval. An agency must explain why a change matters, how it fits the business process and why the commitment is justified. Implementation expertise provides the basis for that explanation, while approval requires communication with the people deciding whether to proceed.

The commitments those decision-makers favor are becoming shorter and more contained. According to the operator, clients seek lower effort, a faster return on investment and minimal lock-in, directing more budget toward practical solutions with demonstrable results and less toward “multi-year transformation plans”. Functional efficiency is also taking priority over growth. Automation, observability, the ability to understand a system through its operating data, and reliability are becoming stronger reasons to invest.

A contained project can still require substantial expertise to define the problem, assess dependencies and establish its business case. That creates a commercial challenge for agencies whose senior people are already carrying more review and client-facing work. Demand for guidance creates an opportunity, while the price a client will accept determines how much expert time the engagement can support. The agency must fit that work within the commitment it sells.

In conclusion

Faster implementation gives agencies more options, but it does not remove the work required to choose, review and approve a useful solution. For agency leaders, capacity planning needs to account for senior judgment and client communication alongside development. Taking on more implementation than those functions can support risks moving delays further along the delivery process.

Commercial decisions need to reflect that wider workload. Smaller engagements can still require substantial expertise to define the problem, assess dependencies and secure approval. Pricing must account for that contribution, including the preparation that makes faster delivery possible. Otherwise, reduced implementation time can leave agencies doing more expert work without a sustainable return.

For executives buying agency services, the strongest proposal connects a contained investment to a clear business requirement. It should explain what will improve, how the result will be assessed and what ongoing costs or dependencies the business will accept. A working prototype helps inform that decision; it does not settle questions about reliability, security or operational fit.

AI capabilities and discovery channels will continue to change. Agencies have a more durable role when they help clients make sound decisions and carry useful work through to operation. The measure of capacity is how much of that work the agency can complete, not how much code it can produce.