By Rachael Ferm, Director of Investments, Florida Funders

Note: Speedrun is the accelerator program by globally leading venture capital firm Andreessen Horowitz (a16z). The highly selective program receives 20K+ applications for ~60 spots each cohort.

While the fall Speedrun cohort surfaced a thesis about expression and judgment, the spring cohort (SR006) focused on the infrastructure underneath that shift.

The common thread weaving this cohort together was not a specific sector opportunity but a structural position: the information layer precedes the judgment layer, and the companies that own the information layer accumulate the context judgment requires.

The companies compressing the data assembly layer are accumulating visibility into how institutions function: where judgment occurs, how expertise propagates, which processes produce outcomes, and which edge cases repeatedly matter.

As AI becomes embedded into daily work, that visibility changes organizations themselves. Judgment becomes more observable, and provenance becomes more valuable. Institutional context increasingly becomes the infrastructure through which expertise scales, oversight moves upstream, and organizational culture, training, and decision-making begin compounding in fundamentally different ways.

Scaling Judgment

Organizations rarely lack expertise, but rather accessible context.

A commercial real estate lending workflow that takes 90 hours is not 90 hours of underwriting judgment. Most of it is retrieval, reconciliation, coordination, and institutional memory scattered across systems, PDFs, inboxes, and people.

Historically, scaling expertise meant hiring more employees. But in many industries, hiring more people dilutes expertise. Meanwhile, context compounds it.

The opportunity is not simply capturing institutional context of the best talent, but making it reusable across workflows. This ability to scale the top quartile of an enterprise breaks traditional economics of organizational structures. 

This is especially important in industries where operational knowledge is concentrated inside aging workforces. In construction, logistics, facilities management, and field services, much of the judgment layer still exists primarily in people: undocumented failure modes, vendor knowledge, sequencing decisions, operational intuition. Retirees take context with them.

AI creates the first economically viable way to operationalize institutional memory at scale. That context is becoming the new moat, whether operational, relational, procedural, or regulatory.

Workflow as Infrastructure

Over time, the workflow wedge becomes infrastructure.

Physical industries make this dynamic easiest to see because the consequences are unusually concrete.

Historically, enterprise software in these industries was sold into centralized IT systems that sat downstream from operational pain. The companies succeeding now are entering much closer to the operational bottleneck itself: the technician sourcing parts, the estimator reconciling fragmented scope documents, the operator managing uptime risk.

That changes the buyer, the pace of adoption, and the role a startup can occupy inside a large institution. Growth rates for pre-seed and seed businesses are skyrocketing. 

For the first time, startups are able to scale into real, operational bottlenecks where labor time and operational risk with physical consequences are already concentrated. A delayed repair, a missed scope item, or a sourcing error already carries a known cost. The product only needs to prove that the information-assembly layer can be compressed without increasing operational risk.

This is important because it captures a key constraint of current AI systems. Models are highly capable at pattern recognition and language generation, but far weaker at modeling persistent environmental state and reasoning forward through physical consequences. Physical environments require systems that understand not only what is happening, but what is likely to happen next under real-world constraints. Understanding the world through physics, and real human interaction, persists as a major barrier to the next phases of innovation. 

In these environments, product impact is evaluated by operational consequence. The machine either returns to service or it does not. The project stays within margin or it does not. That creates unusually strong feedback loops where the operational environment itself becomes a source of grounded training data. Even increasingly capable simulation and synthetic training environments ultimately depend on accurate real-world interaction data, physics, and environmental state transitions to remain useful.

Which means the companies solving workflow problems today are not simply building software, but they are accumulating the operational context future physical AI systems will require. This is critical as we remain in the very early innings of unlocking the physical world, and the massive labor and service budgets it brings along. 

When Judgment Becomes Visible

The same dynamic extends beyond operational systems and into organizations themselves. Once AI compresses the information-assembly layer, institutions begin seeing human judgment differently too.

Historically, organizations relied on outputs and then business outcomes as proxies for judgment. As AI becomes embedded into daily work, organizations increasingly need visibility into how decisions were made, what systems were used, where judgment actually occurred, and which processes produced reliable outcomes over time.

So the important shift is not simply that AI automates tasks, but that it changes how institutions evaluate human activity. 

As the potential reusability of context becomes more powerful, this trend compounds. Companies that discern judgment from process, and reuse that judgment, earn a marked competitive advantage.

As synthetic output becomes cheap, provenance becomes more and more valuable. This shifts organizations toward signals that are harder to manufacture: demonstrated judgment, embedded reputation, trusted networks, observed decision-making over time, contextual trust built inside known systems and relationships.

In Workforce software alone, we saw this from three different angles: Acceler8 executes top-down succession and workforce planning for white collar work, Quinn focuses on workforce training for blue-collar work, while Cedar works on breaking traditional referrals by scaling trusted professional networks. We see how each is a distinctive wedge providing newfound visibility into optimal workforces and career progression. 

This also requires shifting oversight upstream, where tools become involved at the point of idea generation and execution. From a single message leaving the organization to a major decision. This is particularly visible in regulated industries, where post-send monitoring becomes structurally insufficient. But the same dynamic extends into hiring, management, and workforce development. We see companies created from this wedge across verticals. 

So we see that the new dynamics of work attribution and human+AI collaboration will increasingly break the fundamental principles of organizational culture and labor. Different AI systems and workflows will produce different organizational behaviors over time. In that sense, culture itself becomes infrastructural. A company’s AI stack will increasingly shape how companies recruit talent, build their unique brand, and, of course, train and develop humans, tools and judgment within the organization.

Intelligence Commoditizes, Context Compounds

The common assumption is that AI commoditizes software. What SR006 suggested is narrower and more structural: AI commoditizes isolated intelligence while increasing the value of accumulated context. The alpha is strongest where the data underlying this context is hardest to understand and utilize.

Frontier models are increasingly converging on similar capabilities. The strategic question shifts from who has intelligence to who has embedded operational context: the memory, workflows, governance structures, and historical decisions that allow intelligence to operate reliably inside real environments. The enduring companies may be the ones that become the operating memory of their industries.

‍