At Ema, we’ve never believed that AI should just 

—own workflows, own accountability, own the job end-to-end.

That’s why we didn’t build “a better chatbot” or new co-pilot.

—which are intelligent, coordinated, adaptive systems of AI agents that can tackle 

 business workflow end-to-end, whether in customer support, sales and marketing, employee experience—or anything else you can imagine.

These agentic systems keep learning from their current and past environments, get better with time, and collaborate with human teams—so they can focus on work that truly matters.

No more repetitive and rote work for humans. 10x the productivity for companies.

But this kind of intelligence isn’t enabled by the current state of LLMs in enterprises.

EmaFusion™, our proprietary model fusion technology

EmaFusion™ beats the best single models in the market—including GPT-4, Gemini, Claude, and O3—on both accuracy and cost.

It fills a massive gap in enterprise AI by combining the best of 100+ models—because enterprises are overpaying for mediocre results, by using a single model for everything (or using basic model routing).

EmaFusion™ offers a seamless plug-and-play experience, backed by enterprise-grade security and design—powering truly intelligent AI Employees and the future of work.

LLMs are rapidly commoditizing. The top models now cost 

 per million tokens, and new open-source options arrive every week.


costs are still high and performance is inconsistent

What gives? Most companies don’t even realize they are overpaying for subpar results by defaulting to one model—often GPT-4—for every task:

: Summarizing a meeting note or classifying a support ticket shouldn’t cost the same as writing code or answering a nuanced legal question by reasoning across multiple systems and teams.

Underperforming on complex or niche tasks

: No single model works well for all tasks. GPT-4 might write well, Claude could reason better, Gemini might beat them both at multilingual tasks, open-source models can be great for domain-specific use-cases.

The best approach to model selection should be task-specific. Instead, after months spent on implementing a single model, enterprises are left with a large bill and infrastructure that is unsustainable and sub-optimal at scale.


"Each model has its strengths. Some are fast and cheap, others are powerful but expensive. Knowing which models to use when, and how to balance accuracy, costs and latency is a constant struggle for enterprises.” — Surojit Chatterjee, Founder & CEO, Ema Unlimited

But mapping tasks to the right model (or model-mix) is a hard problem to solve too.

Enterprises may manually route tasks to different models, or use 

 strategies, where an ML model or rule-engine attempts to classify prompts and pick a 'best-fit' model.

But even this basic model routing approach fails.

 Real enterprise tasks span multiple subtasks—like data extraction, validation, summarization, and recommendation—which no single model handles well in isolation.

Accuracy, latency, and cost tradeoffs are unpredictable:

 The cheapest model might be too slow or less accurate, or the fastest model might hallucinate despite being expensive. These tradeoffs fluctuate across use-cases.

The best solution isn’t routing each task to any “best-guess” LLM—but about using the intelligence of all the models out there, now and in the future—with enterprise-grade security baked in.

This is how we make the most of the LLM revolution.

The Answer Isn’t Routing—it’s EmaFusion™

EmaFusion™ acts as a task-aware brain that does the thinking behind complex enterprise work. When an AI employee is asked to do something like review a contract, respond to an RFP, or generate a quarterly business review, EmaFusion™ first breaks the task into structured subtasks — things like extracting data, identifying risks, or generating language.

It then uses a self-optimizing system to decide which model is best suited for each step.

For instance, it might use a lightweight open-source model like Mistral for fast text extraction, a fine-tuned Ema model for structured data parsing, and escalate to costly models like GPT-4 only if the task requires deeper reasoning or creative drafting.

Once each piece is completed, EmaFusion™ fuses the outputs into a single, high-confidence result.

 Behind the scenes, it’s a coordinated system of model orchestration, cascading fallback logic, and task decomposition — but to the end user, it just feels like one intelligent AI that gets it right the first time.

We’ve spent months analyzing which models perform best on what tasks, and in what conditions and to what trade-offs, to build EmaFusion™:

When we know the task type—say, summarization or classification—we use taxonomy to route requests to models we’ve rigorously evaluated for that category. This ensures every job starts with the most optimal and informed model choice possible.

When the system’s unsure or dealing with novel inputs, a learned router kicks in. This ML-powered component predicts the best model based on task embeddings, historical performance, and real-time feedback. It enables adaptability and precision when facing new, ambiguous tasks.

Why pay for GPT-4 when a smaller model gets the job done? EmaFusion™ always tries cheaper models first—and only escalates when confidence thresholds aren’t met. This keeps latency low and cost efficiency high without compromising on quality.

However complex the problem, EmaFusion™ breaks it down into subtasks, routes each to the best model, and intelligently fuses the answers.

 EmaFusion™ achieves higher accuracy than any single model, including the top ones like GPT-4, Gemini, Claude, at a fraction of the costs.

 in Action: Saving $$ and weeks, while giving the best answers

Say you’re working with a Fortune-500 client, who drops a 120-page Data-Processing Agreement on Friday at 4 p.m. They want to know:

“Which sections of the 2024 Africa Data-Protection Act govern sending biometric data to Singapore, and can you flag any non-compliant language in our draft?”

Here’s how EmaFusion would solve this (instead of the days spent waiting for multiple teams with tribal knowledge to figure this out):

 EmaFusion™ notices the query is legal, privacy-focused, and risk-sensitive.

 A lightweight model skims the contract, pulling every statute citation and building a quick outline. That answers half the request for pennies.

 Unclear clauses (e.g., cross-border exceptions) trigger a jump to our in-house privacy model. It cross-checks Article 31, Section 12 against the client’s wording and flags two mismatches.

 For one particularly tricky clause, EmaFusion™ pings a top-tier commercial model plus a rule-based legal verifier. They rewrite the sentence and attach the exact statutory language.

The system merges notes, highlights risky text in red, and produces a clean compliance checklist—ready to ship back to the client.

Get answers in 4 minutes end-to-end, not four billable hours.

Cost stays in the “small-model” range because the expensive giant only handled 3 sentences.

Legal team walks on Monday morning with a polished, statute-linked memo—no weekend scramble required.

That’s EmaFusion™ in action: start cheap, escalate smart, fuse everything into a single, confident answer your team can trust.

EmaFusion™ was tested on scores of such 

, across finance, legal, healthcare, R&D, customer support, and it outperformed single-model baselines in terms of 

 — higher than GPT-4o and Claude Sonnet, which score ~91.7%

+17 points accuracy over GPT-4; +6 points over O1 and +9 points over top open source players like Deepseek-R1

 than top-tier models due to intelligent model selection and caching

EmaFusion™’s advantage isn’t just cost savings—it’s getting 

more reliable, high-quality outcomes across diverse enterprise workflows

EmaFusion™ combines outputs from over 100 models—private, public, domain-specific, open-source, including the latest releases like DeepSeek—so your AI investments are always optimized 

Ema's data governance redacts sensitive information before passing it on to public LLMs. Enjoy compliance with leading standards like SOC 2 Type I & II, ISO 27001, 

, HIPAA, GDPR, NIST CSF—and get unbeatable security with top-tier encryption and customizable, private models.

 Our customers can always control which models should be in the consideration set... You can also BYOM (bring your own model)

and easily add to the model garden of EmaFusion™, so you can achieve the same quality in enterprise specific contexts.

Our customers can always control which models should be in EmaFusion™’s consideration set. You can also BYOM (bring your own model)

You can use EmaFusion™ out-of-the-box—no custom fine-tuning or model chaining required. Just plug EmaFusion™ into your stack through our API, or use Ema’s built-for-humans UI. EmaFusion™ readily integrates with 

200+ of the most common enterprise apps and systems, such as

 Salesforce, ServiceNow, SAP, Workday, Snowflake and more.

The future of enterprise work isn’t hundreds of tools or brittle integrations. It’s 

—by deploying a system of AI Employees that deeply understand business workflows and collaborate with human teams.

But for these AI Employees to work well, they need a brain that doesn't compromise—on accuracy, speed,or costs—at scale. 

Instead of forcing enterprises to choose, EmaFusion™ offers a new path: a 

future-proof, self-optimizing LLM architecture

 that’s optimized for how businesses really work.

It’s one of the most strategic, technical, and rewarding bets we’ve made at Ema.

Read our full research paper on arxiv now: 

EmaFusion™: A Self-Optimizing System for Seamless LLM Selection and Integration

How EmaFusion™ beat GPT-4 at 1/20th the cost: Peek into the Brain of our Universal AI Employees 

Bigblue is a leading European logistics platform that empowers 500+ e-commerce and retail brands with seamless fulfillment and delivery solutions. As their customer base grew, so did the complexity of handling support queries—ranging from shipment tracking, order delays, to returns and delivery failures. Managing these inquiries became increasingly challenging, especially as Bigblue scaled its operations and worked with multiple carriers, suppliers, and various stakeholders in the logistics and delivery funnel.

Bigblue’s Challenge: Scaling Customer Support Without Compromising Experience

In the absence of Bigblue, e-commerce brands would have to manually tie up with local carriers to ship their products and manage returns, delivery status updates, and product concerns. But Bigblue takes care of all of this, coordinating with carriers, e-commerce brands, warehouses, and various stakeholders in the delivery funnel, so that their merchants can succeed.

This makes Bigblue’s mission to scale customer support all the more challenging:

Growing, unpredictable volume of support tickets:

 As Bigblue expanded—acquiring new customers and partnering with more e-commerce vendors and carriers—the number of support requests surged. Depending on the time of the year, around festive or holiday seasons, there are also spikes in incoming customer tickets. But relying on additional staffing to handle this growth quickly becomes cost-prohibitive and operationally taxing, underscoring the need for a more 

scalable, cost-effective, reliable approach.

Bigblue’s customers may face a range of concerns. In case of issues arising from delayed or lost shipments, or damaged or incorrect products, a specific response is needed. Each situation demands a nuanced approach, requiring the verification of the claim and a resolution that meets the customer’s needs. This requires coordinating with multiple disparate stakeholders to gather the right information, carefully analyzing and acting on it.

 Another challenge for Bigblue is managing a Europe-wide customer base that submits inquiries in various languages, including English, Spanish, and French. On top of that, the company adheres to strict linguistic guidelines for its support agents—for instance, using the informal “tu” instead of the formal “vous” in French—to ensure culturally appropriate communication.

 Bigblue partners with more than 50 carriers, hundreds of e-commerce brands, and numerous warehouses. Additionally, much of the customer information would be stored in CRMs and internal databases. For support to scale effectively, it must seamlessly integrate with each of these platforms, ensuring the right information can be retrieved to resolve every ticket efficiently.

The Solution: Ema’s Customer Support AI Employee

Ema’s Agentic AI-powered AI Employees gave Bigblue the flexibility, reliability, and scalability they needed to 10x customer support, ensuring nuanced responses in multiple languages with human-like empathy and quality at scale—so much so that some of Ema’s answers are now being used as templates for helping human agents be more efficient and accurate.

Ema’s Customer Support AI Employees continuously learn from past support cases and all relevant documentation, adapting its responses in real-time to stay aligned with evolving delivery and shipment details. It intelligently categorizes and routes tickets, retrieves relevant SOPs, and seamlessly integrates with CRMs, carrier websites, and internal databases to synthesize accurate resolutions.

Unlike static bots, Ema engages in dynamic, multi-step conversations, adjusting its responses based on evolving context. Beyond answering queries, it takes action—coordinating with stakeholders, updating records, and even closing tickets once resolved—delivering 

truly autonomous, high-impact customer support:

: While Bigblue's customer care team on average responds to customers within 2 hours , Ema enables them to deliver accurate and nuanced responses in less than 90 seconds.

Efficient Handling of Fluctuations in Business Cycles:

 Cyclical businesses often experience surge in customer queries in peak seasons. Ema helped Bigblue maintain its SLAs with high customer satisfaction during the peak holiday seasons, without the operational hassles of hiring and onboarding a temporary workforce.

: Immediate, accurate responses to common queries improved the overall customer experience, reflected in better CSAT (Customer Satisfaction) scores.

: Ema rapidly learned to craft structured responses to customer queries, which then served as the gold standard for training new human agents to deliver top-tier customer support.

: Ema’s analytics layer can surface recurring pain points, whether delays at certain routes or issues with specific courier partners, giving leadership the data they need to refine processes.

Impact: Faster, Smarter, More Scalable Support

By leveraging Ema’s Customer Support AI Employees, Bigblue has reduced response times, operational costs, improved customer satisfaction, and freed up human agents to handle more complex, nuanced cases.

The ability to resolve repetitive queries instantly and accurately at scale, while maintaining brand voice and multilingual accuracy has given Bigblue an edge in logistics-driven customer support. With AI-powered Customer Support, Bigblue now delivers a superior, scalable support experience that keeps pace with its rapid growth.

"Ema integrates seamlessly into our ecosystem, autonomously resolving inquiries in seconds across multiple languages. The result has been greater efficiency and smoother operations for our merchants. We’re excited to deepen our integration and partnership with Ema Unlimited.” — Laetitia, Customer Care Team Lead

Ema’s AI Employees are designed to 10x customer support by delivering instant, accurate, and multilingual responses—without compromising on quality. 

 today and transform your support operations.

How Bigblue Uses Ema for AI-Powered Customer Support 

, the latest addition to our Customer Experience Suite—an integrated suite of powerful AI employees designed to redefine support operations with Agentic AI.

Unlike traditional chatbots that rely on FAQ searches and predefined scripts, Ema goes beyond answering questions—she resolves customer issues autonomously, end-to-end. With the ability to take complex actions, adapt to conversations dynamically, and maintain your brand voice, Ema consistently delivers natural, human-like interactions with your customers.

Ema is already helping leading enterprises automate more than 75% of support interactions in most cases while maintaining CSAT scores above 80%1. By elevating customer experience, Ema isn’t just reducing support costs—she’s turning cost centers into strategic revenue drivers.

Ema has been a game-changer for us! We’re seeing nearly 98% accuracy in responses, and it now handles 80% of our 150,000+ monthly tickets. Plus, our CSAT has jumped to 70-75%—a huge improvement over the 40-50% we typically see in FinTech.

–Venkatraman Narayan, Moneyview’s Head of Customer Experience

The Problem: Why Customer Experience Is Still Broken

Customer expectations have never been higher—they demand instant, personalized, 24/7 support. The quality of customer support is what decides if users will return to and love your brand.

Yet, despite increased spending, companies struggle to meet SLAs and retain customer trust. Here’s what’s holding them back:

 – High-pressure, high-turnover roles lead to burnout, longer wait times, and lower CSAT.

– Manual call reviews cover only 3-4% of interactions, leaving leadership blind to real-time quality.

Incomplete Knowledge Bases, Inconsistent Answers

 – Poorly documented knowledge makes it hard to deliver accurate, consistent responses.

 – Limited to simple queries on chat, they fail when customers deviate from scripts or require multi-step actions, leading to lost trust and higher support costs.

Ema’s Customer Experience Suite: Reimagining the Entire Contact Center

Customer experience isn’t just a support challenge—it spans the entire contact center, from Q&A to knowledge management, insights, and automation. Fragmented solutions lead to inefficiencies, disconnected workflows, and missed opportunities.

At Ema, we’ve built AI Employees that optimize the full CX value chain. Unlike siloed tools, our AI employees share data, learn collectively, and collaborate in real time, creating a continuous improvement loop that drives better results.

The best part? You can start anywhere and then expand seamlessly as your needs grow. With Ema’s Customer Experience Suite, businesses shift from reactive support to proactive customer success.

Ema has been a game-changer for us at Envoy Global. With its unparalleled accuracy and efficiency, it has enhanced our approach to the various elements of customer experience.

–Mahi R. Inampudi, CTO, Envoy Global and President, Envoy India

How Ema’s Customer Support AI employee Transforms CX

Ema delivers end-to-end resolution in natural language, optimizing every customer interaction for brand loyalty and business growth. Here’s how:

 – Understands customer intent, context, and policies to provide personalized, dynamic responses across channels and languages.

Autonomously resolves queries by performing multi-step actions like interacting with third party vendors, placing orders, or understanding an image. The best part? Ema’s agentic AI ensures you don’t have to pre-define every single workflow.

Intelligent agent assist (human in the loop): 

Knows when to escalate and provides real-time recommendations to human agents for faster, more effective support.

 – Grounded in enterprise data and policies, ensuring transparency, compliance, and adherence to brand guidelines. Continuously learns from admin and end user feedback.

 Learns from past tickets, eliminating the need for scripted workflows, replacing them with simple natural language instructions that can easily be maintained by non-technical users.

With Ema, support teams shift from answering every query to supervising AI-driven resolutions, stepping in only for complex cases. Here’s more on how each capability works.

Ema delivers natural, fluid interactions in 150+ languages, maintaining brand consistency and policy adherence. She integrates with enterprise systems like CRMs to personalize responses and recognize returning customers to deliver a tailored experience.

She keeps conversations on track even when customers switch topics, provide unclear details, or ask complex questions, maintaining context and responding with clarity and empathy.

Beyond just answering questions, Ema detects frustration and urgency, adjusting her tone to reassure customers and provide the right level of support. She also knows when to step back—whether due to business policies, sensitive topics, or the need for human intervention.

True resolution isn’t just answering questions—it’s executing the right actions across multiple systems. A simple refund request may require verifying order history, checking eligibility, updating records, triggering cancellations, notifying finance, and confirming resolution with the customer. Missing a step leads to frustration and repeated contacts.

Traditional systems rely on rigid, predefined workflows that fail when customers switch topics or introduce unexpected scenarios—leading to dead ends, repeat interactions, and escalations.

Ema takes a smarter approach. Powered by 

, she understands customer intent, dynamically plans actions, and orchestrates multi-step processes. She gracefully handles failures, missing data, and edge cases while ensuring compliance with business policies.

With Ema, support teams no longer need to script every scenario manually. They define key objectives in natural language, and Ema intelligently adapts—delivering faster, more personalized, and scalable customer support. Here’s a comment from one of our 3PL customers, who utilizes Ema’s multi-turn conversation and action capabilities:

Logistics is complex as agents need to coordinate with third-party vendors and make real-time decisions based on data on the ground. With Ema, we’ve achieved a 4.8/5 rating across multiple categories and languages on these complex issues. In many cases, its responses surpass human agents and are now part of our Agent SOP.

– Head of Customer Experience, e-Commerce Fulfillment Provider

Intelligent assist with human in the loop

As your AI employee, Ema balances autonomy with human collaboration. She knows when to handle issues herself and when to escalate—routing cases to the right agent based on expertise, past interactions, and workload.

When an agent steps in, Ema shifts into Agent Assist mode, summarizing conversations, surfacing insights, and suggesting the right next steps. She retrieves order details, policy guidelines, and even drafts responses, so agents can focus on resolution, not research.

Ema is your teammate. She learns from top agents, scales their expertise, and helps the entire team work smarter, faster, and more effectively.

Ema gives admins full oversight and control, ensuring transparency while adapting to your business needs.

 Every interaction is fully auditable. Admins can review her responses, understand her reasoning, and refine her behavior to align with company policies.

 Ema continuously improves by learning from both user interactions and admin feedback, adapting in real time.

Support leaders can fine-tune her tone and personality to ensure she reflects the brand’s values. 

 Beyond standard reports, Ema’s metrics dashboard offers deep insights into customer issues and knowledge gaps. The best part? Admins can take actions, right from within the dashboard.

Enterprise support is messy—issues evolve, and not everything is neatly documented. Ema doesn’t need rigid inputs; she learns directly from your helpdesk.

Just connect your help desk, and she learns from past tickets—identifying patterns and understanding how your team resolves issues across chat, email, and voice with minimal setup.

Updating her is as simple as using natural language—no complex training, no manual rule-setting. Ema is built for easy deployment and maintenance, keeping your support process agile and efficient.

Ema is built to meet industry-leading compliance standards, including GDPR, SOC 2 (Type I & II), ISO 27001:2022, HIPAA, and NIST. Our data governance framework ensures PII is redacted before being passed to public LLMs, maintaining strict privacy controls.

With top-tier encryption and customizable private models, our comprehensive security approach reflects our commitment to handling sensitive data with the highest level of care.

Explore Ema’s Customer Experience Suite

Unlike other vertical solutions, Ema is a horizontal Agentic AI platform whose suite of AI employees automate every aspect of 

 — from autonomous deflection to agent assistance, QA, knowledge management and product insights.

Further, with Ema’s Generative Workflow Engine™, you can create custom AI employees that can tackle the most complex workflows in any business function, amplifying work across teams.

Hire Ema to elevate your customer experience

Customer experience is no longer just about answering questions—it’s about taking action, making decisions, and delivering real outcomes. Businesses embracing fully agentic AI are shifting from reactive support to proactive, autonomous customer success.

With Ema, companies are automatically resolving over 75% of issues with 98% accuracy and 80%+ CSAT.

Ready to Transform Your Customer Experience? See how leading brands are using Ema—check out our 

Introducing Ema’s Customer Support AI Employee: Make Every Customer Interaction Extraordinary

Ema Unlimited is pleased to share this blog from Everest Group authors, Vaibhav Bansal, Vershita Srivastava, and Yusuf Ahsan.

Agentic AI and its reliance on foundation models

The AI and automation space has been evolving at an unprecedented pace. A couple of years ago, Generative AI (gen AI) reshaped our understanding of AI’s potential. Just as the world was beginning to adapt to gen AI, agentic AI then emerged as the new kid on the block and captivated everyone’s attention. While gen AI was about knowledge assistance, agentic AI takes it to the next level through goal-based action, autonomy and adaptability

Though agentic AI is highly disruptive, it is not an isolated new technology. It builds on the capabilities of earlier innovations, especially that of gen AI and foundation models. The foundation models essentially act as the brain for agentic AI, facilitating input analysis, intent recognition, goal setting, advanced reasoning, workflow planning and decision making. This continuous loop of perceive-reason-decide enables agents to perform a range of tasks in varying environments.

DeepSeek has shaken the foundation model ecosystem

Since the launch of ChatGPT in Nov 2022, Open AI has been the industry leader for foundation models. There have also been challenges to that dominance from competitors such as Anthropic, Google, and Meta. All this while certain norms were established by these players around the model ecosystem, such as the need for advanced chipsets, high training costs, significant usage costs and the effectiveness of larger models.

DeepSeek’s entry has challenged these norms like never before. Early tests indicate its superior performance vis-a-vis competitors as depicted in Exhibit 1.

Moreover, this comes at a fraction of the cost of the existing models as outlined in Exhibit 2.

While a few claims may be debatable, DeepSeek has presented a strong proof of concept and disrupted the industry. The question now though is how such limited resources have delivered such a superior offering?

Let’s dive a bit ‘deep’ to ‘seek’ the answers:

DeepSeek’s R1 utilizes multiple expert networks, with only a few activated at a time. For any given query, DeepSeek R1 utilizes only 37 billion parameters out of the total 671 parameters, resulting in a reduced computation cost

It operates on a combination of shared experts with broad capabilities and specific experts with narrow focus areas. A built-in load balancing mechanism ensures even distribution of tasks enhancing efficiency and resilience.

It utilizes a multi-stage training pipeline which combines reinforcement learning and supervised fine-tuning as opposed to relying only on supervised fine-tuning which inflates cost.

Traditional models generate responses one token at a time, with each prediction depending on the previous one. DeepSeek predicts multiple tokens in parallel improving processing speed without major accuracy trade-offs

With these techniques, DeepSeek has challenged the dominance of the incumbent model providers who were winning owing to their compute capacity and scalability. So, how will these advances impact the broader foundation model ecosystem?

: With DeepSeek’s entry, other model providers will be forced to reduce their usage cost. These have been coming down gradually since ChatGPT was launched but expect that to accelerate going forward. Even if a direct cost cut doesn’t happen, more intelligent models will be available at the same price.

: DeepSeek utilized H800 chips by Nvidia instead of the latest H100 GPUs, which the majority of its competitors utilize. Despite the overall demand for chips continuing to climb, the product mix will lean away from the higher end GPUs, eroding manufacturer’s profit margins and reducing hardware costs for end users in the process.

: DeepSeek’s approach could be a differentiator in the short run, but will cease to be a moat in the longer run. As other model providers catch up, the techniques adopted will get standardized with foundational models becoming commoditized.

: Open-source models provide the opportunity to build, modify and integrate the offering as per requirements. While open-source models have found good adoption earlier, DeepSeek is accelerating the creation of an open-source community.

: With Explainable AI, DeepSeek prioritizes transparency and insights on how it arrives at the conclusion. This fosters a trust in DeepSeek and will push others to open their models similarly.

There has been a lot of excitement around Agentic AI’s potential to transform businesses and enable the shift towards a combined digital-human workforce. However, there have also been headwinds. The most common is the concern around cost, contributed by the high usage cost of proprietary models. Also, with usage-based pricing becoming the norm, enterprises are increasingly unprepared for uncertainty around foundation model bills.

Another challenge is the Agentic AI infrastructure. Most organizations are unable to invest in their own infrastructure to host models. Hence, they consume these models through cloud services, which subjects them to data security risks and becomes detrimental in regulated industries.

Concerns around the lack of trust in the outputs generated by an AI agent, pushes back the whole narrative around autonomy.

The disruption caused by DeepSeek is expected to solve these challenges to a good extent. Exhibit 3 demonstrates the expected short-term impact of foundational model commoditization on players involved in an agentic value chain.

While there might be variation in the extent of impact on individual players in the short term, the overall agentic AI market is expected to get a boost in the longer term. Let’s understand how:

Cost reduction, availability of open-source models and greater explainability will lead to a surge in adoption. As Jevons paradox kicks in, enterprises will be able to infuse more processes with agentic AI, without compromising their bottom line leading to increased overall spending. Agentic AI will get more democratized, expanding its reach to a range of enterprises, beyond those with significant financial resources.

While foundation models have received all the attention, the real value for enterprises is at the application layer. With commoditization, the development of domain-specific AI agents will accelerate and with the supply-side becoming more competitive, ultimately this will offer more choice and better quality to enterprises.

DeepSeek has shown us the ‘art of possible’ and might mean reduced short to mid-term profitability for foundation model players. We can expect this to be a trigger for further model engineering, optimization and innovation with advances on domain-specific and multi-modal models. Further breakthroughs in this field are going to benefit the entire agentic ecosystem boosting adoption

Enterprises that continue to rely on just one foundation model for all their needs are going to be at a continuous disadvantage going forward. The enterprise demands today are so complex and diverse that a single model can’t fulfil all of them, while being accurate as well as cost-effective. With new innovations like DeepSeek taking place every few weeks and the goalpost for the best or the most superior model constantly changing, enterprises can’t keep on switching their entire agentic ecosystem from one model to another. Therefore, model fusion technologies which dynamically combine or leverage different foundation models, are expected to become more prevalent. These technologies will not only optimize model cost and accuracy for enterprises, but will also make their agentic ecosystem future-proof by providing the flexibility to incorporate new innovative models seamlessly.

DeepSeek has provided the answer to the most prominent concerns plaguing enterprises in their agentic journey. However, these are by no means completely addressed. Enterprises, especially in regulated or data sensitive industries, are grappling with data privacy and security concerns, which DeepSeek may not have done much to alleviate. In fact, the exposure of DeepSeek’s data including chat history and sensitive information has aggravated this concern. DeepSeek’s evasive nature when encountering questions related to controversial national issues has put into question its reliability in the face of censorship.

Nevertheless, DeepSeek has expanded the horizon through innovation. With a slew of advances expected to follow, models getting more and more commoditized and model fusion becoming a norm, this is the perfect opportunity for enterprises to embark on their agentic journey. The barriers to entry have been brought down significantly allowing enterprises to sail ahead on their agentic journey without a heavy financial luggage.

All posts that contain tag Automation Systems

Evolution of the Polyglots: Where Enterprise Search, Automation Systems and LLMs Fall Short