Enterprise assisted-coding is the new Mega Drive. How to survive the creative chaos?

There is a meme circulating that compares Claude Code to the Sega Mega Drive: how professionals now rush home from work to keep coding, the way children once rushed home to keep playing Streets of Rage. The comparison captures something real about the current moment: people are building things for the pleasure of building them, and the tools have become fast enough (and forgiving enough) that an evening of work can yield something that actually runs live with enough functionalities, authored by someone who was never able to go that far before.

Claude Code reached an estimated $2.5 billion annualised run rate by early 2026 (SemiAnalysis 2026). In a February survey of fifteen thousand developers by the Pragmatic Engineer, it was named the most-used AI coding tool, surpassing GitHub Copilot for complex tasks and earning the highest share of "most loved" responses at 46 percent. Four percent of public GitHub commits are already authored through Claude Code, and projections from the same SemiAnalysis report suggest that figure could exceed twenty percent by year's end. Meanwhile, at Epic — the healthcare technology company behind MyChart — over half of Claude Code usage comes from non-developer roles: support staff and implementation teams who adopted it in ways the company had not anticipated (VentureBeat 2026).

Doomsday SaaSpocalypse

Alongside the assisted coding popularity, a familiar apocalyptic narrative has emerged. The so-called SaaSpocalypse wiped roughly $2 trillion in market capitalisation from software companies between January and February 2026, as investors recalculated what per-seat licensing is worth when AI agents can replicate the work those seats were doing (DigitalApplied 2026). Salesforce shares fell around 33 percent. Atlassian reported its first systemic decline in enterprise seat counts. Commentators have declared the end of software-as-a-service with the confidence usually reserved for predictions that never quite arrive on time.

The doom narrative, however, misreads where the disruption is headed.

A Retool survey of 817 builders published in February 2026 found that 35 percent of enterprise teams have already replaced at least one SaaS tool with a custom build, and 78 percent expect to build more proprietary tools over the coming year. So yes, some tremor is coming.

No sandbox: it’s a construction site

Sixty percent of respondents reported building software outside of IT oversight entirely. Crucially, 64 percent of those shadow builders were senior managers and above. These are not rogue interns experimenting in a sandbox, but experienced operators choosing to construct their own solutions because they can now do so in days rather than months (Retool 2026).

Marketplace spring x Enterprise spring

This is the detail that most commentary overlooks. The conventional prediction says: the cost of building software has dropped massively, therefore we will see a flood of new applications on the market, something like the early App Store explosion of 2008–2012.

That prediction follows a clean logic, and it is mostly likely wrong. Or rather, it describes a secondary effect while missing the primary one. The main current of the coming flood is not arriving on a public marketplace. It is arriving inside companies, in the form of proprietary tools, internal dashboards, custom plugins, and bespoke workflow automations that never leave the organisation that created them.

LLMs reduce the marginal cost of software creation and maintenance, allowing enterprises to internalize workflows that were previously outsourced to SaaS. The resulting advantage is not the existence of custom tools (we had those), but the continuous co-evolution of software, processes, and proprietary data. Over time, this creates path-dependent operational systems that are difficult to replicate externally, not because of technical barriers alone, but because they embed organization-specific knowledge and iteration history.

The SAP lesson

Bill Vass, CTO of Booz Allen, offered a historical framing in a recent interview that makes the structural point clearly. Before ERP systems became widespread in the 1990s, most large organisations ran on custom-built internal software. For example, tailored exactly to how the Army wanted it, or Bank of America, or BMW.

Then came the harrowing consolidation wave: SAP, PeopleSoft, Oracle. These enterprise vendors became powerful oligarchy, creating legacies in terrible UX that persist until today in large and low-cost-driven organizations. SAP is a great example of that, deemed as the worst user experience of all time.

The promise was standardisation and cost reduction; the price was flexibility and fit. Every business person, Vass recalled, hated it (Tech Brew 2026). The new generation of AI-assisted coding tools, he argued, could swing the pendulum back toward individual customisation — not because everyone has become a developer, but because the cost structure that made customisation prohibitive has fundamentally changed.

The implication for SaaS companies is therefore more nuanced than the apocalyptic framing suggests. Salesforce is a behemoth. It will not vanish. The structural advantages of systems of record (deep databases, network effects, regulatory compliance infrastructure, integration ecosystems) remain formidable.

Gartner estimates that 35 percent of point-product SaaS tools will be replaced by AI agents by 2030, but this pressure falls unevenly.

The tools most at risk are the smaller, single-function applications: the $20,000-a-year workflow automations, the niche admin panels, the internal help desks that a competent operations lead can now rebuild on a platform like Retool in an afternoon.

At Harmonic, a startup discovery platform, one automation lead hit a breaking point with exactly this kind of tool, rebuilt it internally, and triggered a cultural shift: the company now runs 33 internal applications, and when someone proposes buying new software, the default question has become whether they could just build it instead (Retool 2026).

Plugin marketplaces already exist, of course: Figma has one, Salesforce has one, HubSpot has one. But until now they have been largely confined to a developer niche, populated by teams with the technical capacity to build and maintain integrations. The shift currently underway is that this capacity is spreading across business functions. According to one analysis of vibe-coding communities, 63 percent of active users in early 2026 are non-developers (product managers, founders, marketers, operations leads) building full-stack applications and internal tools in natural language (Solveo via Grey Journal 2026).

The YC Winter 2025 cohort offered a startling data point: 21 percent of companies reported codebases that were more than 91 percent AI-generated. When even startup founders treat code generation as a commodity input, the economics of who builds what — and where that building happens — change in ways that reach well beyond the developer community.

The TL;DR about marketplaces is that more free plugins are likely to emerge, be reused, and further developed. Which is again not great news for small, medium and niche SaaS.

Seniors, leads and managers: what is at stake?

The more interesting consequence, at least for the readers of this newsletter, is what happens to professional roles when building becomes a distributed activity rather than a specialised one. If a product manager can prototype a working internal tool in a day, and an operations lead can wire it into Salesforce and Slack with proper access controls, the developer's role shifts.

It does not disappear, though. The Retool data is instructive here: 51 percent of builders have shipped production tools, but moving from prototype to production still requires someone who understands security reviews, role-based access, data integrity, and the architectural decisions that determine whether a fast build becomes a durable system or an expensive piece of technical debt.

In practice, this means developers are becoming something closer to coaches, reviewers, and system architects for a much broader base of builders. They provide the guardrails, the final polish, the governance frameworks.

This is not unlike what happened to graphic design after Canva: the skill did not vanish, but it migrated upward in the value chain, toward the problems that require judgement about systems, style or cohesion rather than execution of individual outputs.

The BCG Nordic AI report published earlier this year makes a related point about organisational design: successful AI initiatives require dedicated cross-functional teams that combine business, operational, data, and technology capabilities with shared accountability for results (BCG 2026). The building itself becomes a collaborative act, not a handoff from "business" to "engineering" but a continuous loop where domain expertise and technical oversight coexist.

Creative chaos and inherent risks

The risk, naturally, is sprawl. Sixty percent of the builders in the Retool survey were operating outside IT oversight. Only 19 percent of organisations described themselves as having advanced AI automation maturity. Thirty-seven percent have not yet established any AI productivity metrics at all.

The shadow IT problem of the 2010s (when every department ran its own unsanctioned SaaS subscriptions) can easily repeat itself in a more dangerous form: unsanctioned custom applications, built quickly, connected to production data, with no audit trail and no governance.

The difference between a Mega Drive and enterprise software is that nobody's production database was at stake when you left Sonic paused overnight.

For strategists, designers, and product managers, the operational question is not whether building will become more distributed. The Retool data, the adoption curves, and the pricing pressure on SaaS vendors all suggest it already has. The question is how organisations structure the governance, the skill development, and the collaborative workflows that make distributed building durable rather than chaotic.

The companies that get this right will not be the ones that build the most tools. They will be the ones that build the right tools, maintain them, and know when to stop building and start buying.

---

Sources cited:

• SemiAnalysis (2026), "Claude Code is the Inflection Point"
• Pragmatic Engineer Survey, February 2026 (15,000 developers)
• VentureBeat (2026), "Anthropic says Claude Code transformed programming"
• DigitalApplied (2026), "The SaaSpocalypse: AI Agents Disrupting Software Industry"
• Retool (2026), "The Build vs. Buy Shift" (survey of 817 builders)
• Tech Brew (2026), "Is it really the end of SaaS as we know it?"
• Grey Journal (2026), "Best Vibe Coding Tools in 2026, Ranked" (citing Solveo analysis)
• BCG (2026), "The Nordic AI Inflection Point"
• Gartner estimates on point-product SaaS replacement (via Intellectia 2026)