The Influencer Series is an intimate, invite-only gathering of influential, good-energy leaders. The intent is to have fun, high-impact, “dinner table” conversations with people you don't know but should. The Influencer Series has connected over 4,000 participants and 15,000 influencers in our community over the last decade.
These roundtable conversations provide a space for prominent VC funds, corporate leaders, start-up founders, academics, and other influencers to explore new ideas through an authentic and connective experience.
Most organizations approach AI with expectations of a single, all-powerful agent to solve every problem. This monolithic approach creates inefficiencies in enterprise computing, with models too slow for simple queries and too expensive for routine tasks.
Jeffrey Akiki, co-founder of Not Diamond, challenges this convention by advocating for networks of specialized AI agents. His counterintuitive vision suggests that less sophisticated components working together can outperform single, more powerful systems.
This article explores how specialized agent networks can revolutionize enterprise AI through intelligent routing, modular architecture, and meta-layer optimization that delivers superior performance at a fraction of the cost.
Specialized AI agent networks outperform monolithic models while reducing costs by up to 10x and improving response times significantly.
Intelligent routing systems direct queries to appropriate specialized models, similar to Google's revolutionary approach to search architecture.
Modular networks provide enterprises with transparency and control, enabling gradual adoption paths unlike black-box AI solutions.
Meta-layer optimization through intelligent routing creates competitive advantages surpassing even the most advanced individual AI models.
Many organizations approach AI with an almost mythological expectation – they envision a single, omniscient agent that can magically transform any input into the perfect solution. When applied to actual enterprise computing environments, this mindset creates more problems than it solves.
In complex organizations, the most sophisticated tasks rarely succeed through individual effort. Instead, they require different types of expertise working in concert. A customer service inquiry might need both technical knowledge and emotional intelligence. A product development cycle combines market research, engineering, and user experience design. AI systems should mirror this natural division of labor.
Large, monolithic models face inherent limitations when handling diverse enterprise requests. They're often too slow for simple queries, too expensive for routine tasks, and too rigid to adapt to specific business contexts. It's like trying to use a sledgehammer for both demolition and delicate carpentry, and these one-size-fits-all solutions prove inefficient and potentially destructive.
Instead of pursuing the myth of a single supreme AI, Jeffrey Akiki advocates for a more nuanced approach: orchestrating networks of specialized agents, each precisely calibrated for specific tasks within a broader intelligence ecosystem. This architecture creates a spectrum of capabilities, from fully autonomous systems handling defined tasks to collaborative assistants augmenting human work.
Picture a modern hospital, where specialists collaborate to deliver comprehensive patient care. Similarly, AI agents can excel in focused domains – some perfect for rapid document retrieval, others built for deep data analysis, and still others optimized for nuanced customer interactions. When properly coordinated, these specialized units deliver superior results compared to generalist systems.
In practice, this means deploying smaller, purpose-built agents for routine requests. These lightweight systems process standard queries at a fraction of the computational cost while matching or exceeding the performance of their larger counterparts. For more complex challenges, requests can cascade through multiple specialized agents, each contributing their unique expertise to construct comprehensive solutions.
The beauty of this approach lies in its scalability, efficiency, and effectiveness. Like a well-organized team, each agent focuses on what it does best, creating a sum greater than its parts.
The evolution of search engines offers a powerful lesson for AI architecture. In the early days, Yahoo attempted to be the "everything website" – a single destination for weather, sports scores, news, and more. Google took a radically different path, building a sophisticated router that connected users with the most relevant information sources.
At Not Diamond, this principle drives the development of intelligent routers that automatically direct queries to the most appropriate specialized model. The router becomes a meta-intelligence, making split-second decisions about whether a simple FAQ-trained model can handle a query, or if the situation demands a more sophisticated reasoning agent.
This architectural choice creates inherent resilience. When individual components need updates or improvements, they can be modified without disrupting the entire system. The router continues directing traffic while specialized agents evolve independently, much like how Google's search infrastructure adapts while individual websites come and go.
Within enterprise computing environments, the benefits of specialized routing manifest in three critical dimensions. Response times plummet when simpler models handle routine queries, avoiding the computational overhead of larger systems. Let's face it – a basic customer inquiry shouldn't require the same processing power as complex analysis.
Cost efficiency skyrockets as organizations deploy expensive, sophisticated models only when their advanced capabilities truly add value. Not Diamond's testing reveals up to 10x cost reductions while maintaining or improving overall performance quality. Their default router, trained on extensive evaluation datasets, consistently outperforms individual state-of-the-art models through intelligent model selection.
Beyond incremental advances, these improvements represent a fundamental shift in how AI systems can deliver value. To break free from traditional technology trade-offs, organizations that match the right tool to each task can simultaneously enhance speed, reduce costs, and improve quality.
Customer support exemplifies the power of specialized AI networks. Many organizations currently deploy sophisticated models for every query, regardless of complexity – the equivalent of using a supercomputer to add single-digit numbers.
Basic questions about return policies, order tracking, or company FAQs don't require advanced reasoning capabilities. Lightweight models, trained on company FAQs, can handle these queries with remarkable speed and precision. When customers face more complex scenarios, like damaged items requiring special handling, the system seamlessly engages more sophisticated reasoning models equipped to navigate nuanced policies and exceptions.
Behind the scenes, the intelligent router orchestrates this dance between different AI capabilities. Customers experience consistent, high-quality support while the organization optimizes its resources, creating a rare win-win in enterprise technology.
Large organizations often hesitate to deploy probabilistic AI models in production environments. Their concerns aren't unfounded – when something goes wrong with a black-box system, finding the root cause becomes a nightmare of debugging and uncertainty.
Modular agent networks flip this equation. When issues arise, teams can quickly isolate which component failed and why. This granular visibility aligns perfectly with enterprise requirements for controlled testing and systematic evaluation. Organizations can start small, testing individual components before expanding to more comprehensive deployments.
The architecture inherently supports the kind of gradual, controlled adoption that enterprises require. Each module can be independently verified, monitored, and improved, creating a path to innovation that doesn't require betting the business on unproven technology.
While tech headlines focus on the latest releases from OpenAI and Anthropic, a quieter revolution is unfolding. Hugging Face hosts nearly a million models, showcasing unprecedented democratization of AI capabilities. This isn't following the cloud computing playbook, where a few giants eventually dominated the landscape.
Instead, AI is experiencing remarkable divergence. Fine-tuning models and developing specialized solutions becomes more accessible every day. Rather than a few "shark" models consuming everything in their path, we're seeing an "army of piranhas" – countless specialized models creating a richer, more adaptable intelligence ecosystem.
Organizations can now break free from vendor lock-in, building custom intelligence layers using their proprietary data. This shift fundamentally alters the competitive landscape, creating opportunities for innovation beyond the reach of even the largest AI providers.
What gives companies a real competitive edge in AI is the orchestration of intelligence at the meta-layer, beyond merely having the biggest models. Here's the thing: Not Diamond's evaluations reveal a surprising truth. Their routing system outperforms individual state-of-the-art models, including heavyweights like GPT-4 and Claude 3.5 Sonnet, on standard benchmarks.
When organizations intelligently route queries to smaller models like GPT-4 Mini as appropriate, they can achieve superior performance while dramatically reducing costs. The integration of custom models, fine-tuned on proprietary data, creates truly differentiated capabilities that can't be easily replicated.
This meta-layer optimization represents a new frontier in AI architecture. The winners won't necessarily be those with the most powerful individual models, but those who most effectively coordinate specialized intelligence to solve real-world problems.
As we look toward the horizon of AI development, Jeffrey Akiki's perspective offers a compelling alternative to the conventional wisdom. The future isn't about creating digital demigods that can do everything, but rather about building intelligent ecosystems where specialized models collaborate seamlessly to achieve superior results.
His parting advice to aspiring AI entrepreneurs – "Build and not plan as much" – encapsulates both the urgency and the opportunity in this rapidly evolving field. While the possibilities for specialized AI seem boundless, the key to success lies in creating systems that can effectively harness and coordinate these diverse capabilities.
Investing globally since 2001, BASF Venture Capital backs startups in Decarbonization, Circular Economy, AgTech, New Materials, Digitization, and more. Backed by BASF’s R&D and customer network, BVC plays an active role in scaling disruptive solutions.
A premier international law firm with deep expertise in Corporate Venture Capital, WilmerHale operates at the nexus of government and business. Contact whlaunch@wilmerhale.com to explore how they can support your CVC strategy.
FinStrat Management is a premier outsourced financial operations firm specializing in accounting, finance, and reporting solutions for early-stage and investor-backed companies, family offices, high-net-worth individuals, and venture funds.
The firm’s core offerings include fractional CFO-led accounting + finance services, fund accounting and administration, and portfolio company monitoring + reporting. Through hands-on financial leadership, FinStrat helps clients with strategic forecasting, board reporting, investor communications, capital markets planning, and performance dashboards. The company's fund services provide end-to-end back-office support for venture capital firms, including accounting, investor reporting, and equity management.
In addition to financial operations, FinStrat deploys capital on behalf of investors through a model it calls venture assistance, targeting high-growth companies where FinStrat also serves as an end-to-end outsourced business process strategic partner. Clients benefit from improved financial insight, streamlined operations, and enhanced stakeholder confidence — all at a fraction of the cost of building an in-house team.
FinStrat also produces The Innovators & Investors Podcast, a platform that showcases conversations with leading founders, VCs, and ecosystem builders. The podcast is designed to surface real-world insights from early-stage operators and investors, with the goal of demystifying what drives successful startups and funds. By amplifying these voices, FSM supports the broader early-stage ecosystem, encouraging knowledge-sharing, connectivity, and more efficient founder-investor alignment.
Alchemist connects a global network of enterprise founders, investors, corporations, and mentors to the Silicon Valley community.
Alchemist Accelerator is a global venture-backed accelerator focused on accelerating seed-stage ventures that monetize from enterprises (not consumers). The accelerator invests in enterprise companies with distinctive technical founders and provides founders a structured path to traction, fundraising, mentorship, and community during the 6-month program.
AlchemistX partners with forward-thinking corporations and governments to deliver innovation programs worldwide. These specialized programs leverage the expertise and tools that have fueled Alchemist startups’ success since 2012. Our mission is to transform innovation challenges into opportunities.
Join our community of founders, mentors, and investors.