One thing has become evident in the early months of 2026: the industry is no longer in triage mode. We’re seeing a structural reallocation of human and machine capital – not just a rationalization of headcount, but a recomposition of how work gets done. Block’s recent decision to slash its headcount by 40% is a prime example. CEO Jack Dorsey framed the move not as a response to financial distress, but as a proactive pivot toward "intelligence tools" and leaner, more agile teams. February's selloff in Cloudflare and other cybersecurity incumbents following Anthropic's Claude Code Security launch is the kind of market signal we watch closely – AI is beginning to compress the value of point solutions that were, until recently, considered defensible.
These developments suggest that the market is not merely optimizing for efficiency – it is establishing a new baseline for operational velocity and efficiency.
I recently spoke with one of our best engineers, and what he described is a precise illustration of how human capital is being restructured in real time, not in the abstract. Rather than managing a team, he became one: spinning up a Redis system for memory, launching an MCP (Model Context Protocol), and deploying a fleet of specialized agents to execute specific tasks in parallel. He organized them as "directors" across five engineering configurations, spinning up nine agents each to handle specific rules, tasks, and production. This engineer effectively created the equivalent of 45 engineers overnight, completing 50 tickets – the kind of work that would have taken a traditional team weeks or months – in less than two days.
Another company in our portfolio built its architecture specifically to rip and replace its codebase every six months. They realized early on that what we build today evolves so fast that we cannot remain tethered to holistic, monolithic environments. They are designing for a reality where "subject matter expertise" is no longer a barrier to entry.
In this new reality, the tried-and-true codebase is an anchor. Work that once commanded a two-week sprint is now a two-hour task. The standard for technical excellence has shifted from durability to disposability. AI models are rapidly eclipsing average human capability for standard development tasks.
Put simply, the era of "steady growth" is over. The market has entered a period of radical Darwinism where the traditional startup playbook focused on incremental engineering shifts and legacy hiring is no longer just obsolete; it is a liability. For a startup to survive, it must undergo a fundamental cognitive shift. This isn't just about moving fast. It’s about a total reimagining of product velocity, efficacy, and the very nature of a competitive moat. For example:
- Opendoor recently launched an end-to-end mortgage product in six weeks – a feat that traditionally required two years.
- High-level software development is shifting from writing code to orchestrating agents. At Cursor, the shift is tangible: 35% of their merged pull requests are now generated autonomously by agents in cloud environments.
- Decagon rejected the assumption that a single general-purpose model could meet the latency and accuracy demands of enterprise CX and instead built a network of specialized models, each post-trained for a distinct function and further widening the gap between what the major labs prioritize and what production environments actually require.
The real advantage is the ability to create micro-organizations or communities of agents working within your company to build tools and write scripts. This creates massive underlying leverage. While not everyone can run this immediately, organizations can be designed to enable individuals to manage multiple agents. This allows companies to quantify the additional leverage gained from their current engineering team – creating a literal multiplication factor in the number of agents per engineer – and rethink their entire organizational design.
The Block case is not an isolated event – it is a bellwether. We are moving toward a world where we don't necessarily need to hire more; we need to enable the "more" within a smaller, elite core.
This creates an unprecedented opportunity for the agile. We are witnessing the birth of a new class of investable companies – entities that leverage AI not just as a tool but as a source of subject-matter expertise integrated directly into their workflows. This will create a new generation of founders who can launch companies faster and with less capital. Unemployment in 2030 will be resilient because these tools will allow people of all ranks and skill sets to build and launch their own businesses, evolving far beyond just VC-backed startups into a whole new realm of company creation.
One strategic implication that the most forward-thinking of these companies are already wrestling with is around outsourcing the heavy lifting to foundation model providers. However, this calculus has a shelf life. Open-weight models are closing the capability gap faster than most expected and the frontier is no longer the exclusive domain of a handful of API gatekeepers. The most competitive app-layer companies will not remain pure consumers of other people's models indefinitely. The ones that build durable moats will be doing increasing amounts of post-training and model work themselves, including fine-tuning, RLHF, domain-specific adaptation, and treating the model layer as a product decision, not a procurement one.
However, a surprising point of agreement between AI optimists and skeptics remains: the "delegation of thought" cannot yet be fully deferred to machines. This doesn't mean agents lack the potential for superior efficiency, but they still lack the high-level context of a specific business environment. The role of the principal engineer is shifting – it is no longer about manual execution but about defining priorities, integrating disparate systems, and managing the strategy that AI cannot yet replicate autonomously. The human element cannot yet be removed.
The future of engineering and product isn't necessarily about how teams code, but about how they actually manage agents that write the code. It's that human element that's going to help protect what gets deployed and enhance the sophistication of its application to businesses. Here are three things worth pressure-testing as founders navigate this moment:
- Audit architecture for disposability, not durability. If codebase or organizational structure is built to last, that may be a liability. The companies winning right now are the ones that have made it structurally easy to rip, replace, and redeploy. Founders should ask themselves “how long would it take my team to rebuild the core product from scratch with today's tools?”
- Recalculate hiring math before the next headcount decision. Founders should be looking at what the agent-to-engineer leverage ratio looks like on the current team. The question is no longer "do we have enough people?" but "are we generating enough from the people we have?" Companies that can quantify this will operate at a fundamentally different cost structure than those still defaulting to traditional team scaling.
- Double down on your principal layer. The most durable competitive advantage right now is a small number of people who are exceptional at setting context, integrating systems, and making high-judgment calls that agents can't yet replicate. Founders need to identify who those people are and bear hug them.
- Move like you've never moved before – and if you can't, change your culture first. Speed is no longer a competitive advantage. It's the price of admission. The companies that survive this moment won't be the ones that eventually adopt these tools, they'll be the ones that already have and have restructured their entire operating cadence around them.
The window to restructure around these realities is open now. It won't stay open indefinitely.