Light Versus Heavyweight IT
The changing nature of work
Software and IT consultancies are selling off sharply, amid deep market uncertainty about what the future holds. Naturally, the root cause is AI. Investors fear that AI’s rapid emergence will reduce demand for both enterprise software and consultancy services. In this week’s blog, we try to get a grip on what’s going on. We start off the sense-making with a binary framework.
Heavy versus light
Heavy and light are two IT paradigms for innovation. Heavyweight IT focuses on stable, integrated enterprise systems managed by IT specialists for reliability and security. Lightweight IT emphasizes agile, user-driven tools that enable quick experimentation outside formal IT controls.
Heavyweight IT represents traditional back-end infrastructure: servers, transaction systems, and complex integrations. It prioritizes efficiency, resilience, and business logic, but often stifles rapid innovation due to its rigidity and high replacement costs.
In contrast, lightweight IT spans consumer-grade tech, such as apps, tablets, and IoT devices. In general, these devices are non-invasive, presentation-layer solutions built by users or third parties. Workers are often using these tools on their own desktops as a replacement for clunky enterprise UIs. These tools foster innovation through cheap, flexible deployment, bypassing IT departments for immediate needs.
The two are complementary: lightweight IT relies on heavyweight platforms for data and stability, while heavyweight platforms strengthen over time from lightweight innovations.
AI is lightweight
AI apps like Anthropic’s Claude and OpenAI’s Codex unfold within the lightweight IT paradigm. These tools are agile, consumer-grade interfaces accessible via web or CLI, and don’t require IT department approval. Claude handles reasoning, code explanation, and document processing with simple prompts, while Codex generates code from natural language with minimal setup.
Their lightweight nature shines through in non-invasiveness: no deep backend integration, just presentation-layer augmentation for quick tasks like prototyping or refactoring. Right now, they are mostly user-driven tools, but in some settings, their use can be supported by management.
Individual workers propel workplace adoption. Surveys show 37% of knowledge workers use AI regularly, while 68% experiment despite barriers, such as access policies. Employees under 40 and college-educated lead (34-40% usage), sharing tips via Slack or demos to spread “behavioral contagion.” This shadow IT dynamic removes structural hurdles, boosting productivity 40%+ in prototyping via multimodal inputs, before formal rollout.
Software sinks into the infrastructure
AI apps like Claude and Codex drive enterprise software into the infrastructure, morphing it into the heavyweight IT paradigm. Traditionally, software apps played the lightweight role across enterprises as agile, user-facing tools integrated via APIs, middleware, and data pipelines into rigid back-end systems. In the future realm, AI will gradually take over as the primary user-facing tool. Alas, it is the AI agents that use software based on automation or human prompts, not by interacting with humans themselves.
Once plugged in, enterprise software becomes opaque infrastructure, prioritizing stability over flexibility. AI agents handle routine tasks, such as querying databases, automating workflows, and using outputs from one software app as inputs in another to accomplish tasks faster. The core enterprise software becomes invisible to users who interact via blackboxed AI UIs. Thus, user-driven lightweight AI tools are the catalyst, turning dynamic software into stable plumbing.
Implications for software
Management won’t chuck out enterprise software anytime soon. These systems have evolved over decades, embedding a million custom reports, endless debugging, and software updates. Companies have tailored their workflows around how the software works. So, ripping them out risks chaos. Compliance adds ironclad lock-in: large companies with bureaucratic layers demand audits, data sovereignty, and vendor accountability that AI overlays can’t yet fully replicate without heavy certification.
However, the light-to-heavy transformation will reshape enterprise software. AI apps will commoditize some user interfaces, simultaneously as specialized SaaS remains vital: we have core logic, data conversion (Adobe’s PDF monopoly), and workflow orchestration (Salesforce ServiceNow) that endure as indispensable plumbing. Blackboxed AI queries these via APIs, yet human oversight, compliance, and edge cases demand robust UIs and vendor lock-in.
Sinking into the infrastructure will likely change SaaS pricing models over time. Exactly how is difficult to stipulate now, it’s early days. However, there is a clear and present danger that per-user pricing will crumble as individuals swap bloated interfaces for universal AI overlays, slashing logins and feature bloat. Customers might prefer a pivot to outcome-based billing (e.g., per query, per task) or emphasize volume over subscriptions. Over the long haul, shifts like these are likely to negatively influence revenue growth for names like Salesforce, ServiceNow, and Adobe.
In any event, software incumbents try to embed their own AI overlays to retain relevance. If these efforts are successful, they can mitigate margin erosion. Still, even though these overlays might be very useful, there is also a clear and present danger that customers want to consolidate their AI vendors and replace these tools with Anthropic Work or another all-purpose AI application.
Nevertheless, this dynamic will play itself out over the next decade. Where we end up is impossible to say now.
The changing nature of work
So far, tech workers have seen the biggest impact from AI. What is going on in this realm can entail challenges that more and more categories of workers have to face.
In tech, AI is rewriting the nature of work by fusing roles. One superworker can now handle tasks that earlier required teamwork. A single early AI adopter can roll designer, programmer, and product manager into one entity by leveraging Claude for UI mocks, Codex for code generation, and natural language prompts to spec features end-to-end. Boundaries blur in this kind of workflow: the PM sketches wireframes, debugs APIs, and A/B tests in hours, not weeks. Naturally, productivity surges enormously when you have these kinds of superworkers in your company
Yet AI’s token-hungry nature demands ruthless efficiency. Each query burns compute credits, like employing an extra hire, making casual use uneconomic. Enterprises might soon start limiting access to their top performers; for with these kinds of costs, you’d better be a superworker to defend the outlays. Paradoxically, this can save software’s role as the core user interface for most knowledge workers. Companies can only afford to give the most experienced and talented access to the holy grail of AI. It can also mean that marginal employees risk obsolescence, as AI amplifies skill gaps, and the cost of computing falls.
A recent study found that AI sparks overwork. Workers accomplish 2-3x more, compressing projects and inflating expectations, 60-hour weeks become the norm, and burnout looms. “Accomplishment addiction” sets in: finish faster, take on more.
Time compression hits IT consultancies hardest. AI shrinks project timelines dramatically. What once demanded months of team toil, from requirements gathering to deployment, now prototypes in days via a lone consultant wielding Claude for specs and Codex for code. Utilization craters as billable hours evaporate; a $200/hour senior resource idles while juniors scramble for scraps. As a consequence, there is a high likelihood that time-based billing, the industry’s lifeblood, will implode. So, what can IT consultancies do to protect their sales levels? This is difficult to answer now. Pricing models could change based on outcome-based results, fixed-fee transformations, or high-margin AI governance (fine-tuning, integration, ethics guardrails). What is more certain is that the billable body-count model slowly dies. Those who want to survive within the business must come to terms with the fact that the future belongs to the orchestrators, not the assemblers.
Will this yield mass white-collar unemployment, or unleash demand for skilled orchestrators? History tilts toward the latter: tech waves absorb labor into new roles. Yet if AI tokens cheapen more slowly than capabilities grow, displacement could sting. The jury’s out.
Disclaimer: Important Information for Retail Investors
The information in this blog is for educational purposes only, not financial advice. Investing in stocks carries risks; past performance doesn’t guarantee future results. Conduct thorough research and seek advice from financial professionals before investing.
The author is a retail investor, not a licensed advisor. Due to changing market conditions, content accuracy isn’t guaranteed. All investments have risks, including the potential loss of principal. Assess your risk tolerance and goals before investing; diversification is key to managing risk.
The author may have positions in the mentioned stocks, which can change without notice. Readers should do their due diligence and consult professionals before acting on blog information.
Before investing, verify information from credible sources, understand prospectuses and financial statements, be aware of your financial situation, and consult professionals for aligned investment choices.
Readers are responsible for their investment decisions; the author is not liable for any outcomes. Investing in individual stocks carries risks; therefore, exercising caution, conducting thorough research, and seeking professional guidance are recommended for informed investment decisions.