AI agents upend work and SaaS: from code-writing bots to the death of management

The leap from 'AI-first' to 'AI-native' is not just a matter of swapping out old tools for smarter ones—it’s a wholesale reinvention of how organizations think, operate, and structure themselves. As Prukalpa Sankar puts it, being AI-native means AI is 'integrated into the fabric of our thinking, operations, and processes,' requiring companies to reimagine workflows from first principles rather than simply layering AI onto legacy routines. This transformation is evident in hiring practices at companies like Atlan, where candidates must now demonstrate AI curiosity and skill—sometimes through live AI prompting in interviews—to ensure every new team member is aligned with the AI-native mission and ready to leverage agentic tools from day one.

AI-native transformation is fundamentally altering organizational charts and team dynamics, with companies like Atlan and Ramp leading the charge by embedding AI agents directly into daily operations. At Atlan, for example, a single customer success manager now orchestrates up to nine specialized AI agents, each handling discrete sub-tasks such as call intelligence, dramatically boosting productivity and responsiveness. Similarly, Ramp reports that 50% of its code is now written by AI, and non-engineers are empowered to ship production code, signaling a shift where traditional roles are redefined and the boundaries between technical and non-technical work are blurred.

This agentic revolution is not limited to engineering or product teams; it is reshaping entire organizations, from go-to-market and finance to customer support and beyond. Miro, for instance, has automated labor-intensive processes like competitive intelligence and board preparation with AI agents, while AT&T’s orchestration of 'super agents' and domain-specific 'worker' agents has driven productivity gains of up to 90% across a workforce of over 100,000 employees. The result is a new paradigm where humans increasingly act as supervisors and strategic decision-makers, overseeing fleets of AI agents that execute the bulk of operational work.

However, the journey to AI-native is as much about culture and leadership as it is about technology. Companies like Block and Zapier emphasize that transformation fails if leaders don’t use AI tools themselves, and that immersive exercises—such as hackathons where engineers must rely solely on agents—are critical for surfacing both the excitement and discomfort that come with redefining roles. As agent capabilities rapidly advance, organizations must invest in upskilling, foster trust in AI outputs, and rethink management itself, with some, like Ramp, declaring 'management is probably dead' in favor of optimizing individuals as empowered builders alongside their digital coworkers.

Agentic AI is fundamentally democratizing software development, enabling non-technical users—from marketers to founders—to build functional applications through natural language prompts and intuitive interfaces, a phenomenon popularized as 'vibe coding.' Platforms like Lovable, Emergent, and Google's AI Studio exemplify this shift, boasting millions of non-developer users and valuations soaring to billions, while case studies show solo founders and marketing teams rapidly shipping production-ready apps without writing a single line of code. This surge is not just a Silicon Valley curiosity: Emergent, for example, claims over 7 million user-created apps in just eight months, and OpenAI’s internal data agent lets 4,000 employees analyze petabytes of data using plain-English queries, signaling a global redefinition of who can build software and how quickly it can be done.

This democratization is radically reshaping the role of developers, shifting their focus from manual coding to orchestrating, overseeing, and refining the work of AI agents. As coding agents generate the bulk of new code—Ramp reports 50% AI-generated code, aiming for 80%, while Andrej Karpathy claims he now programs 'mostly in English'—developers are increasingly responsible for architectural decisions, quality assurance, and managing multi-agent workflows. The new developer workflow resembles an assembly line or 'factory farming of code,' where orchestration, system design, and judgment replace syntax mastery, and the bottleneck shifts from implementation to defining valuable problems and ensuring robust, maintainable outputs.

However, this rapid acceleration comes with new risks and responsibilities: AI-generated code is often more error-prone—studies show up to 41% more bugs and 30% higher logic error rates—necessitating vigilant human oversight, robust review processes, and the emergence of 'harness engineering' to constrain and guide agentic outputs. Enterprises and startups alike are grappling with verification bottlenecks, technical debt, and the challenge of maintaining code quality at scale, as highlighted by incidents like Replit's accidental codebase deletion and the need for strict architectural discipline in vibe coding environments. As Simon Willison warns, 'true AI-assisted coding requires reviewing and understanding AI-generated code,' underscoring that while the cost of implementation plummets, the value and scarcity now lie in oversight, system design, and the ability to manage AI at scale.

The rise of agentic AI has also sparked a wave of new paradigms and workflows—such as multi-agent orchestration, 'prompt requests' replacing pull requests, and closed-loop development—requiring both technical and non-technical users to develop new skills in system design, context engineering, and collaborative oversight. Tools like Replit, Google AI Studio, and Morph are racing to provide infrastructure that supports parallel agent-driven development, modular sub-agent architectures, and seamless integration with existing workflows, while companies like Ramp and StrongDM are pioneering frameworks to onboard every employee into AI-powered building. As the competitive advantage shifts to organizations with the highest density of 'sovereign builders,' the future of software creation is less about typing code and more about orchestrating agents, shaping outcomes, and managing an ever-expanding ecosystem of AI collaborators.

AI agents are fundamentally disrupting the traditional SaaS business model by shifting value creation away from seat-based licensing and proprietary software toward usage- and outcome-based pricing, domain-specific application, and user data ownership. Companies like Cursor and Cognition exemplify this trend: despite relying on third-party foundation models from OpenAI and Anthropic, they are highly valued ($20bn and $10bn, respectively) for their ability to leverage user data and develop embedded models that differentiate their offerings and reduce dependency on external model labs over time. This transition is forcing SaaS vendors to innovate in pricing and value capture strategies, as their cost of goods sold (COGS) is increasingly tied to external model pricing, limiting their control over both performance and profitability, and challenging the defensibility of traditional seat-based models.

The rise of AI agents is accelerating a market-wide shift from seat-based SaaS pricing to usage- and outcome-based models, as automation reduces the need for human seats and buyers increasingly evaluate software based on outputs rather than headcount. Major enterprises like Publicis Sapient are already slashing SaaS licenses by up to 50% in favor of AI tools, while industry analysts predict that by 2028, 70% of SaaS vendors will have refactored their pricing around consumption or organizational capabilities. Leading vendors such as Salesforce, Intercom, and Notion are experimenting with hybrid models—charging for seats while layering on AI usage fees—but these are seen as temporary stopgaps as the market moves toward models that more closely align pricing with the value delivered by AI-driven automation.

AI agents are also reshaping competitive dynamics and defensibility in SaaS by lowering switching costs, commoditizing the application layer, and enabling new entrants to challenge incumbents with counterpositioned business models. As AI agents orchestrate workflows across siloed systems and automate complex tasks, legacy SaaS products risk being relegated to mere data storage layers, while startups like Decagon and Dyna.Ai adopt value-based or outcome-driven pricing that incumbents like Zendesk struggle to match without cannibalizing their own revenue. This has triggered a wave of market rationalization, with investor confidence shifting toward data infrastructure and security platforms, and a $285 billion wipeout in software stock valuations following the launch of autonomous AI agents like Anthropic's Claude Cowork. In this environment, defensibility increasingly hinges on proprietary data, domain expertise, and the ability to deliver measurable business outcomes rather than on user lock-in or feature differentiation.

To survive and thrive amid this disruption, SaaS vendors are being forced to rapidly adapt their go-to-market strategies, product architectures, and pricing models. Companies like Notion, Box, and Canva are embedding AI agents directly into their platforms, shifting toward usage-based and outcome-driven products that deliver tangible productivity gains and business intelligence. Meanwhile, the rise of AI-native startups and platforms that empower non-technical users to build custom apps—such as Emergent, which has enabled the creation of over 7 million apps in eight months—signals a democratization of software creation and a move toward more personalized, agent-enabled solutions. As a result, the SaaS landscape is fragmenting into a mix of hybrid pricing models, with vendors experimenting with subscriptions, usage fees, and even ad-supported tiers to capture value across diverse user segments and maximize revenue in a world where software is increasingly driven by AI agents rather than human users.

Enterprise-scale adoption of agentic AI is fundamentally constrained by the immaturity of governance frameworks and the complexity of integrating AI agents into existing workflows. Even industry leaders like Google and Replit have struggled, as Amjad Masad points out, because enterprises often treat agents like traditional software rather than recognizing the need for a 'fundamental rethink and reworking of workflows and processes.' The messy, fragmented nature of enterprise data—structured and unstructured, scattered across silos—compounds these challenges, making reliable deployment and effective crawling a formidable task.

The security paradigm for agentic AI is being upended as agents require broad, dynamic access to enterprise resources, rendering traditional perimeter-based and least-privilege models inadequate. As Mike Clark of Google Cloud observes, 'Security perimeters have been drawn around everything—but that doesn’t work when agents need to be able to access many different resources,' raising urgent questions about how to define and enforce access in a 'pasture-less, defenseless world.' This shift necessitates robust, resource-intensive controls such as testing-in-the-loop, verifiable execution, and strict development isolation, especially in the wake of high-profile incidents like Replit’s AI coder wiping a company’s entire code base.

Cultural and operational mismatches further complicate enterprise adoption, as agentic AI operates probabilistically while enterprises are structured around deterministic processes. Mike Clark notes, 'We don't know how to think about agents... We don't know how to solve for what agents can do,' highlighting a governance mindset lag. As a result, successful deployments in 2025 were typically narrow, heavily supervised, and driven by bottoms-up, no-code or low-code initiatives—prototyping in the trenches rather than sweeping top-down mandates.

AI agents are driving a profound productivity revolution across industries, automating complex workflows and enabling both technical and non-technical employees to contribute directly to software creation. Companies like Ramp and Emergent exemplify this shift, with Ramp reporting that 50% of its code is now written by AI and Emergent empowering users to build over 7 million apps in just eight months. This democratization of software development is compressing timelines dramatically—'8 days of research in 8 minutes'—and shifting workforce composition toward upskilling in AI tool usage rather than traditional coding, as seen in the rapid adoption of low-code/no-code platforms and the rise of AI-native startups growing three times faster than their SaaS peers.

However, these gains come with significant risks and societal challenges, particularly around workforce displacement, upskilling, and the evolving human-autonomy relationship. Large-scale layoffs—such as Klarna's 50% reduction in headcount and over 76,000 global job losses linked to AI—underscore how AI agents are automating roles in engineering, customer success, and middle management, while also threatening traditional SaaS business models as the cost of software creation approaches zero. Yet, human oversight remains essential, with leaders like Mariesa at Pearson and Oji Udezue emphasizing the need for hybrid models, continuous evaluation, and new skills in orchestrating and supervising AI systems to avoid pitfalls like technical debt, trust erosion, and the 'verification bottleneck' in agent-assisted workflows.

The evolving relationship between humans and AI agents is redefining workforce roles from direct task execution to higher-level orchestration, judgment, and system design. As Sam Altman and Ivan Zhao note, the future of work increasingly resembles managing a team of autonomous agents, with humans acting as supervisors, strategic 'pointing' agents, or even 'mayors' of their own digital cities. This shift demands new organizational structures, upskilling in AI fluency—B2B go-to-market roles requiring AI skills have grown 14-fold—and a cultural transformation, as seen at companies like Notion and Nexar, where non-technical employees now build and manage agentic workflows, and management itself is being reimagined to prioritize individual builder excellence.

Despite the promise of AI-driven productivity, integrating agents at scale presents ongoing challenges in governance, trust, and equitable value distribution. Enterprises must invest in robust oversight, permissions, and accountability structures—treating agents like new hires with measurable goals and least-privilege access—to prevent costly errors and maintain customer trust, as highlighted by the 95% of consumers demanding explanations for AI decisions. Furthermore, the societal challenge of ensuring that productivity gains benefit workers as well as companies is becoming acute, with thought leaders like Steve Yegge warning of burnout and advocating for a new balance between output and well-being in an AI-augmented workforce.