Varun is a Product and applied AI leader, heading frontier AI forward-deployed customer experiences, agentic systems, GTM and partnerships.
getty
Frontier models keep getting better and more widely available. That is good news, and it's why the real contest is shifting: when powerful intelligence is something any company can tap, the strategic move is not picking the best model but building the AI-native ecosystem it plugs into.
While many enterprises are experimenting with AI agents, fewer than 10% have scaled them to deliver real value. That gap is an ecosystem problem.
This is the upside of fast-improving models: each release raises the floor for everyone. Leading systems sit a point or two apart, and open-weight options match last year's flagships at a fraction of the cost. Capable intelligence is becoming a shared input any team can build on, so the edge moves to the network of tools, data, identity and distribution that turns it into work.
That network is the ecosystem. In my work building agentic ecosystems, a strong model is necessary but never sufficient; the connective tissue around it decides whether a deployment succeeds.
The Value Has Moved Up The Stack
I believe interoperability is the defining bet of this era. Shared standards for connecting agents to tools and to each other, like Model Context Protocol and the agent-to-agent protocols, are spreading across the industry rather than being locked to one platform. That signals where durable value sits. Owning the pipe matters less than being the most useful, most trusted place on it.
Many platform shifts have followed a similar pattern: as shared protocols mature, differentiation moves higher up the stack. Once the connective layer is broadly available, competition increasingly centers on developer experience, the breadth of integrations, runtime reliability and the governance controls enterprises trust. An AI-native ecosystem is designed with agents as first-class participants rather than treating them as capabilities layered onto existing software.
Why The Ecosystem Compounds
Models keep improving, and many of those gains become broadly accessible over time. An ecosystem is different. It creates a flywheel: more connectors make agents more capable, more capable agents attract more developers and more developers build more connectors. Each turn strengthens network effects, deepens integration and lowers the cost of building new capabilities. While a new model might lift the whole field at once, a well-developed ecosystem compounds its value over time in ways that are much harder for competitors to replicate.
Cost is where this becomes tangible. As agents move into production, token spend stops being a rounding error and starts affecting unit economics. Organizations that scale successfully tend to treat efficiency as an architectural concern: routing routine tasks to smaller or open models, reserving frontier reasoning for the work that benefits from it and reusing context instead of regenerating it. Without that discipline, costs can balloon; with it, organizations can scale agent deployments more sustainably. In that sense, efficiency is becoming an increasingly important source of competitive advantage alongside model capability.
Picture an asset manager running quarterly portfolio reviews across thousands of accounts. One agent pulls positions from custody systems, another gathers market data and a third checks each proposed change against the client's mandate, risk profile and regulatory requirements before a coordinating agent drafts options for an advisor.
Any capable model can sit at the center of that workflow. Much of the defensible value lies in everything around it: integrations with custody, order management and compliance systems; governance controls that define what each agent can do; audit trails regulators expect; and the proprietary data and institutional knowledge that are difficult to replicate.
Organizations that tie that ecosystem too tightly to a single closed stack may sacrifice flexibility as models and standards evolve. Where security, compliance and business requirements permit, building on interoperable standards can preserve more room to adapt over time.
Where The Edge Comes From Now
With intelligence abundant and the connective layer shared, advantage comes from how you position inside the ecosystem, and AI-native organizations are often better positioned to take advantage of that shift. Two moves stand out.
Partnership
As enterprises deploy more agents, orchestrating a growing fleet becomes increasingly complex, and no single vendor can provide every capability. That is where business development becomes strategic. AI-native companies are typically designed to integrate rather than own every layer of the stack, partnering with frontier model providers, specialized tool builders and data platforms where it accelerates delivery instead of rebuilding capabilities in-house.
The payoff is speed and reach: a single integration into a widely used platform can expose an agent to a large installed base far faster than incremental model improvements ever could. Over time, the strongest partnerships do more than expand distribution—they embed a product into workflows other companies depend on, creating ecosystem positions that become more valuable as adoption grows.
Specialization
Rather than competing on raw model quality, many AI-native startups focus on owning a single high-value workflow end to end, becoming the specialized agent or connector that larger platforms and enterprises can plug into. By building on shared standards and frontier models rather than training their own foundation models, they can often reach production faster than organizations assembling every component themselves.
Many also experiment with outcome-based pricing instead of traditional seat-based licensing, allowing them to compete in areas where incumbents' business models are less flexible. That combination of specialization, interoperability and commercial flexibility gives small teams a credible path to competing in markets that once appeared difficult to enter.
The Short Game And The Long Game
In the short term, stay portable and disciplined. Favor interoperable standards where security and privacy allow, so no single vendor's roadmap becomes your cage. Instrument cost from day one, not after the bill arrives. Scope early agents narrowly, where outcomes are measurable and trust is easy to earn.
In the long term, build a position, not just a product. The lasting winners will be the high-value nodes in an ecosystem: the trusted connector, the specialized agent or the tool others depend on. Invest in the partnerships, data and governance that hold up over years, and choose the ecosystem you want to grow within.
Models will keep improving, and that progress will lift everyone. What sets the winners apart is the ecosystem everyone else has to build on. As network effects compound, ecosystem positions become harder to displace, making today's architectural and partnership decisions increasingly consequential.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
>> Home