For the past few years, we have watched artificial intelligence transition from a specialized technological frontier into a commodified utility. AI has bled into our everyday lives, changing how we interact with information. Initially, the market was flooded with conversational tools. More than half of recent Y Combinator cohorts were building AI agents for specific sectors.
But the true change arrived with the advent of mature agentic AI. This evolution granted everyone access to their own autonomous AI agent, capable of retaining context and executing complex workflows independently of any specific industry.
The velocity of this technological development is immense. It is bringing a massive structural change to our society, the future of work, and most importantly, our corporate infrastructure. The fundamental logic of how software is built has changed drastically: the marginal cost of creating software is rapidly approaching zero.
This economic inversion will allow for completely decentralized, bottom-up software building within organizations. Platforms like Lovable are already pioneering this space, giving employees the flexibility to generate and deploy custom applications for the exact micro-use case they need simply by conversing with an AI.
However, this decentralized approach hits a hard ceiling when applied to the enterprise.
The Ceiling of Decentralization and The Generative AI Failure
The profound issue with a purely bottom-up software approach is the absolute lack of quality control, systemic compatibility, and audited scale. For core financial operations, a company simply cannot afford to operate on fragmented, employee-built shadow IT. Wiring millions of dollars, checking and reconciling complex supplier invoices, and managing multi-currency cash flows are processes that require deterministic reliability. You cannot run a corporate finance department on theoretical flexibility.
This tension is exactly why enterprise AI is currently in a state of crisis. A 2025 MIT report, recently highlighted by Fortune, revealed a sobering reality: 95% of generative AI pilots at companies are failing. The research, analyzing hundreds of public AI deployments, stated that the vast majority of these initiatives stall out, delivering little to no measurable impact on the P&L.
Why? Because companies keep trying to bolt new AI features onto inefficient, legacy systems. When the foundational infrastructure is fundamentally broken, adding an AI wrapper doesn't solve the bottleneck; it exacerbates it.
The Legacy ERP Crisis
The alternative to decentralized AI chaos has historically been the traditional Enterprise Resource Planning (ERP) system (typified by Germany's most valuable company, the €223 billion behemoth, SAP).
But legacy ERP systems have been failing at scale. Traditional enterprise software relies on a rigid, outdated assumption: growing companies must adapt their idiosyncratic processes to fit the static parameters of the software. It should be the exact opposite.
The empirical data on this is staggering. Gartner research shows that more than 70% of ERP implementations will fail to reach their original business case goals, and predicts that 25% of such implementations will fail catastrophically. And Panorama Consulting's 2025 ERP report highlights that in complex sectors, implementations fail at a massive 73% rate, burdened by average cost overruns of 215%. Traditional implementations require armies of expensive consultants and force companies to pay exorbitant fees just to customize basic functions for their own needs.
Our Thesis: The Flexibility of Lovable, The Reliability of SAP
We believe that enterprise software must possess the hyper-tailored flexibility of an AI builder like Lovable, but it equally requires the regulatory-grade reliability of an SAP.
That is why we are launching Agent F.
Built in our studio, Agent F is an AI-native ERP system explicitly engineered to challenge the ERP monopoly. We are building a new system from the ground up because it is the only way to act as the ultimate quality control layer in a decentralized software world.
Operating on the core philosophy that "Structure Creates Freedom," Agent F serves as a self-learning central nervous system for the enterprise. Instead of forcing a company into a rigid database schema, Agent F's AI observes, understands, and natively adapts to a company's specific financial workflows. It consolidates scattered tools and spreadsheet chaos into a single source of truth, offering "Conversational Finance" where executives can query their real-time cash position and runway scenarios in plain language, backed by perfectly reconciled, deterministic accounting.
The Team
Executing a thesis this ambitious (challenging a €223 billion titan on its home turf) requires a unique combination of leadership, venture scaling, and technical mastery.
We have assembled exactly that. Our Studio Partner, Julian Teicke, has joined as a Founding Investor to build up Agent F. At the helm driving our daily execution and vision is Felix Schläger, our Founder and CEO.
Validating the absolute technical necessity of this platform, Jürgen Müller (the former Chief Technology Officer of SAP) has joined Agent F as our Technical Advisor. Having led the technology and innovation agenda for the very legacy systems we are disrupting, Jürgen brings a deep understanding of what enterprise-grade scale and reliability truly demand.
We look forward to challenging the legacy ERP market and building the intelligent future of enterprise operations right here in Germany, the very country where SAP was born.
Welcome to Agent F.
Written by Julian Teicke
Chairman
