George Firican

The University of British Columbia

Director, Data Governance & Business Intelligence

Led a team of wonderful and highly skilled people in the areas of:
- Finance
- Human Resources
- Information Systems
- Gift Processing 
- Operations
- Strategic Projects
- Prospect Research & Analytics 
- Data Governance & Business Intelligence 

To support the goals of the fundraising and alumni engagement portfolio across the university.

Chief Advancement Operations Officer (Interim)

Director of Information Technology (Interim)

• Lead the Data Governance/ Quality program and the strategy development of Development and Alumni Engagement’s (DAE) data acquisition, maintenance and dissemination while fostering innovation and long term strategic vision
• Lead the efforts to develop a business glossary as well as a business intelligence (BI) framework and program that provides user-friendly access to the right information for the right people at the right time
• Directed the development and enhancement of standards, procedures, processes, policies related to the improvement and/or maintenance of quality of data within the Blackbaud Enterprise CRM database
• Created and fostered strategic partnerships and influential relationships with DAE senior management, the broader UBC community, campus stakeholders, vendors, and external universities to secure the success of DAE’s data governance, business intelligence and information quality efforts
• Collaborated with Information Technology to manage the development of automated data validations, data cleanses, data quality issue reports, and deployment of new tools and enhancements to aid in the preservation of data quality
• Advanced the data governance and business intelligence profession by participating in professional organizations, acquiring or maintaining required accreditations, publishing articles, and giving presentations at professional events

Associate Director, Data Governance & Business Intelligence

I proudly took part of UBC's "Start an Evolution" campaign (https://startanevolution.ubc.ca/), campaign which raised $1.5 billion dollars for students, research and community engagement and to double the number of alumni involved annually in the life of the university by 2015.

• Established and led an award winning Data Governance/Quality Program to support the university’s $1.5 billion fundraising campaign and sustainable alumni engagement strategy
• Provided oversight of the Data Governance/Quality Program and kept it on track by managing the development of standards, procedures, processes, policies related to the acquisition, management, and dissemination of enterprise data impacting the Blackbaud Enterprise CRM database while leading the Advisory Council
• Developed an award winning Tracing Program through which the returned mail percentage of Development and Alumni Engagement’s Portfolio has been reduced to less than 1%
• Managed the development of business performance metrics and scorecards to monitor and communicate the status and improvement of service of the Records department
• Achieved a 92% address and phone number accuracy for Canadian records, an improvement from ~64%
• Communicated to and educated internal and external stakeholders on the results and initiatives of data governance, quality, master data management, information delivery and advised stakeholders about best practices; was the first and only department to achieve 100% on the annual client satisfaction survey on communication delivery and content

Data Quality Manager

LightsOnData

- Educational Content Creation: Develop and deliver comprehensive courses, workshops, and webinars on data governance, data management, data quality, AI, and more.
- Brand Amplification: Collaborate with data companies to enhance their brand visibility and thought leadership in the data community.
- Engagement and Outreach: Host the Lights On Data Show podcast, featuring industry leaders and experts to share their insights and experiences.

Founder

Pearson

Course Instructor

O'Reilly

iData Quality Academy

Developed course content for iData Quality Academy
- Data Profiling
- Data Quality Costs

Instructor

Self employed

• Provided project management and business analysis expertise to various projects across the different industries (BaniPePost.com, JT Business Group, SMAWS, 8p8, Ashenzi, NetPress Consulting, LTS, AHACommunity, etc).
• Designed and developed unique, highly professional web and print designs, branding and e-learning solutions as well as developing and configuring custom websites (e-stores, online communities, content management systems, blogs, brochure websites, etc.).
• Accumulated experience as a web master by installing, upgrading, performing backups, managing migrations and maintaining a series of complex websites.

Freelance Consultant

Sauder School of Business, UBC

• Served as internal Project Manager for the evaluation, build and deployment phases of a custom-built, scalable web payment gateway (WPG) that complied with the PCI standard requirements and reduced Sauder Finance’s operational time for the allocation and reporting of revenue received via card payments by 75%
• Managed the deployment of a FIPPA compliant e-newsletter tool (Industry Mailout) selected to achieve a standard for design and operational support while reducing expenses by $15k/year
• Planned and managed the development and release of a new service for personal/academic blogging through the use of the WordPress MU environment for faculty members and current Sauder students
• Managed the development of an improved content publishing process to account for 15 digital signs, each with audience specific content & emergency broadcasting capabilities
• Served as the internal project manager for the development and deployment of the Canaccord Learning Commons web presence comprised of a WordPress based website and a BuddyPress student interaction environment
• Developed solutions for integrating the business environment with web 2.0 technologies such as WordPress 3 based sites, online communities (Ning, Moodle), and collaboration tools (MediaWiki, Project Spaces, CoMapping) while providing expertise and system support for these platforms
• Identified customer requirements, analyzed and reviewed product features, integration, security, scalability and performance requirements for projects such as PPM Tool, a collaboration tool to be used between the Center of Health Care Management and BC health care authorities, Desktop Blogging Client Solution, ECM Society Online Presence, Faculty Master Calendar, Form Building Tool, Event Registration, PCI Payment Forms Conversions, and several online presences for different business areas and conferences

Project Manager / BA

Everest iTech

• Promoted to Project Manager for the Showtime Models website and web application and the All Ears accounting software development project due to excellent interpersonal, communication and leadership skills in the relationships with the client and internal development team
• Led the development of several e-commerce sites (ex: SkyHook Solar, Expressive Designs Florist, JT-Autotech) and dynamic websites (ex: Mailinglist System, Informant Zone) and also helped architect their databases
• Increased the employees' knowledgebase by developing e-learning modules and user manuals for the use of the company’s payroll and accounting software
• Enhanced the online awareness of the company’s brand by 40% by creating a new content strategy and redeveloping its website in a WordPress, SEO friendly environment
• Minimized the cost of online advertising by developing free distributed widgets that increased incoming traffic by 5k visitors/ month

Project Manager / Lead Web Developer

Event Technologies

• Developed customer support standards and procedures for the company’s e-store business and helped transition a staff of 12 to this new environment through training sessions and a knowledgebase wiki.
• Minimized the installation time of new systems by 80% by creating and managing standard desktop images based on the employees’ job roles.
• Reduced search time of hardware and software information by 50% by developing a standard for filing all the technical documentation on hardware equipment and software.

IT Consultant

WGOA

• Demonstrated strong leadership skills by acting as the team leader of a software developer group in order to successfully implement and deliver an interactive Go learning application two weeks ahead of deadline and within budget.
• Coded the backend interaction of a merchandise inventory application with a barcode scanner and also contributed to developing the MySQL database and reporting. 
• Facilitated the team members to work remotely by implementing a VPN solution and integrated the use of Subversion to enable software versioning and revision control.

Software developer

💬 Founder of LightsOnData.com and LightsOnData YouTube Channel
Focused on providing informative content such as ONLINE COURSES, templates, guides, best practices, articles, white papers, and other useful resources to help you with your data governance and data management questions and challenges

💬 Creator of online courses that are engaging, new, and thorough. 
Courses: Building an Award-Winning Business Glossary, Data Governance Maturity Models, Data Governance Fundamentals, etc. - check them out here: https://www.lightsondata.com/courses/

💬 Providing LinkedIn content strategy for data-focused companies and products, as well as those with a focus on artificial intelligence (AI), business intelligence (BI, data science, data management, and data governance areas.
Connect with me today and ask how I can help you reach the data audience.

💬 Host of Lights on Data Show - a podcast and LinkedIn Live weekly show that provides insights into the data world and showcases people that help to shape it.

💬 Get some fun data merchandise here: https://shop.spreadshirt.com/lightsondata/all

💬 International key note speaker, conference speaker, and panelist (spoken at over 60 different conferences).

💬 I love being on camera, hiking, surfing, filming, and video editing
YouTube: @LightsOnData


🏛 My experience and specialties: 
• Expertise in data governance, information quality, business intelligence & data management
• Extensive background in web and application development with experience of the full life-cycle of a project
• Strong leadership capabilities, organizational and communication skills

Specialties: 
• Enterprise information management: Information quality, data governance and stewardship, master data management, business intelligence, business process improvement
• Development: Web and application development, relational database design and development
• Planning: Budget and resource planning, agile and waterfall project management, team development and leadership
• Tools and Applications: Collibra, Tableau, MS Visio, MS Project, Premiere, Photoshop

George Firican 💡

Here is my recommendation for this week.
If Agentic AI is on your roadmap, checkout this ebook  👇

It breaks down what many overlook: AI success depends on the maturity of your data architecture.
Here's what else you can find in it:
• Clear explanation of what a modern, AI-ready architecture looks like
• Practical guidance on data quality, cataloging, lineage and governance
• A real example of how a bank scaled agentic AI by modernizing its foundation

So if you are leading data initiatives or planning for AI adoption, this is a good resource to read.

Until next time, let's keep putting the Lights on Data!
-George 
#data #AI #partnerpost

💡 Award Winning Data Governance Leader | Content Creator & Influencer | Founder of LightsOnData | Podcast Host: Lights On Data Show | LinkedIn Top Voice

Modernize Data Architectures for  Agentic AI Success

What are the biggest misconceptions business leaders still have about AI?

I sat down with Stijn (Stan) Christiaens, Co-founder of Collibra, at Data Citizens to unpack what AI readiness really means for organizations. 

Here are 5 takeaways from our chat:

1. Many leaders still see AI as “a chat in the cloud.”
2. Organizations need the right foundations around AI
3. AI ready data requires more than quality
4. Unstructured data is no longer optional
5. Unified governance is becoming essential

If you want a clear picture of what it truly takes to prepare your organization for AI, this interview delivers.
Full conversation in the video below.

Until next time, let’s keep putting the Lights On Data.
-George Firican

#data #datagovernance #AI

Every December I look ahead… and this year, one prediction feels very safe:

I'll be attending Subsurface Live on December 10, coffee in hand, curious about how Agentic AI and open lakehouse patterns will shape 2026.

The agenda lines up well with questions many of us are exploring right now:
➡️ How will agent-based execution reshape data workflows?
➡️ What signals are emerging around performance, formats and multi-cloud patterns?
➡️ Where is the Apache Iceberg ecosystem pushing innovation next?

For anyone designing data systems or planning next-year priorities, this feels like a valuable moment to absorb ideas and future direction.

Registration is free: https://lnkd.in/gmDPiaES
Where: Virtual
Date: Dec 10, 2025 
Time: 8AM - 2PM PT / 4PM - 10PM GMT / 5PM - 11PM CET

See you there!

#data #AI

Greg Coquillo

Product Leader @AWS | Startup Investor | 2X Linkedin Top Voice for AI, Data Science, Tech, and Innovation | Quantum Computing & Web 3.0 | I build software that scales AI/ML Network infrastructure

Product leader with a passion for solving business use cases such as Growth, Go-to-Market, and Price Optimization. I owned P&L for commodity-based products, and leveraged AI to increase the average customer contract values across business industries. As a Product Leader at Amazon, I own a roadmap that drives the adoption of AI products through the Media Automation charter, which supports Private Brands’ YOY growth. I work with Data Scientists, Software Engineers and Business Intelligence Engineers to leverage Machine Learning and Big Data Analytics to design predictive models that enhance business decision-making and performance to accelerate growth, reduce Time to Market of products, monitor cross-geo compliance and import requirements, and ensure customer safety.

I am always curious and eager to learn how new technology affects our daily lives, and I love to support communities on their quest to adopt Artificial Intelligence across industries. Feel free to contact me if you have Data Science projects that you’d like to be showcased on this platform for visibility.

🌐I am a Global Business Leader with 12 years of experience and skills including Product Management, P&L management, Global Networking, EBITDA Optimization, Pricing Management, Margin Management, Product Management, Program Management, Operations and Supply Chain Management, Lean Manufacturing, Six Sigma, and Continuous Improvement.

🧘🏽‍♂️Soft Skills include: Communication, Collaboration, Lateral Thinking, Adaptability, Inspiration, Innovation, Empathy, Listening, Diplomacy, Organization, Responsibility, Leadership...

🎬🎥🎞 I really enjoy watching movies, and if I could redo it all over, I’d be a Sci-Fi script writer so I can be bounded by the limits of imagination.

(Opinions are my own)

Product Management, Media Automation

Owner of the AI vision and roadmap for Private Brand’s Compliance Division. Driving adoption of new technology, allowing my customers to scale their processes and increase productivity. Core responsibilities include:

-Building cross-functional teams consisting of Data Scientists, Software Engineers, Data Engineers, Product Managers and Program Managers, to plan and implement sophisticated AI projects
-Manage the New Initiative Intake Process
-Gather Business requirement artifacts
-Communicate roadmap definition and prioritization
-Coordinate User Acceptance Testing
-Manage Launch Announcements
-Prepare and communicate annual planning artifacts

Understanding vector databases is essential to deploying reliable AI systems.

People usually think “picking a model” is the hard part…
But in real production systems, your vector database decides your speed, accuracy, scalability, and cost.

This visual breaks down the most popular vector databases:

- Pinecone
Great for large-scale search with low latency and effortless scaling. Perfect for production-grade RAG in the cloud.

- Weaviate
Mixes vector search with knowledge-graph structure. Ideal when you need semantic search plus relationships in your data.

- Milvus
Built for billion-scale AI workloads with GPU acceleration. The choice for massive enterprise systems.

- Qdrant
Focused on precise filtering and metadata search. Excellent for personalized recommendations and structured retrieval.

- Chroma
Simple, lightweight, and perfect for prototypes or local RAG setups. Fast to start, easy to integrate with LLMs.

- FAISS
A high-performance library from Meta - not a full DB, but unbeatable for similarity search inside ML pipelines.

- Annoy
Great for read-heavy workloads and fast nearest-neighbor lookups. Popular in recommendation engines.

- Redis (Vector Search)
Adds vector indexing to Redis for ultra-fast queries. Ideal for personalization at real-time speed.

- Elasticsearch (Vector Search)
Combines keyword search with dense embeddings. Useful when you need hybrid retrieval at scale.

- OpenSearch
The open-source alternative to Elasticsearch with vector capabilities. Good for teams wanting full transparency and control.

- LanceDB
Optimized for analytics-friendly vector storage. Popular in data science workflows.

- Vespa
Combines search, ranking, and ML inference in one engine. Large recommendation systems love it.

- PgVector
Postgres extension for vector search. Best when you want SQL reliability with RAG capability.

- Neo4j (Vector Index)
Graph + vector search together for context-aware retrieval. Ideal for knowledge graphs.

- SingleStore
Real-time analytics engine with vector capabilities. Perfect for AI apps that need both speed and heavy computation.

You don’t choose a vector database because it’s “popular.”
You choose it based on scale, latency, cost, and the type of retrieval your AI system needs.

The right database makes your AI smarter.
The wrong one makes it slow, expensive, and unreliable.

Most people think an LLM “just answers your question.”
But under the hood, 15 complex steps happen in a fraction of a second each time you type a prompt.

This visual breaks down what actually goes on inside a transformer model—from raw text, to numbers, to meaning, to the final generated answer.

Here’s what’s happening behind the scenes:

🔸 Your Text Becomes Data
Your prompt is captured exactly as written, cleaned, normalized, and broken into tiny pieces called tokens.

🔸 Tokens Become Numbers
LLMs don’t understand words - only numerical representations.
Each token is mapped to an ID and then converted into dense embeddings that carry meaning.

🔸 Context & Structure Are Added
Positional encoding tells the model the order of the words, and multi-head attention analyzes grammar, relationships, and dependencies from many perspectives.

🔸 Deep Reasoning Happens Internally
Through multiple transformer layers, the model figures out:
– What you’re asking
– What information matters
– What the likely next word should be

🔸 Your Output Is Generated Step-by-Step
The model predicts the next token (not the whole sentence), evaluates probabilities, samples the best continuation, and converts tokens back into text—repeating this loop until your answer is complete.

This is the hidden engine behind every AI conversation:
math + probability + massive pattern recognition, all running continuously until the model is confident in its response.

If you want to understand LLMs deeply, this is the foundation.

Do you think Gemini is the only big AI release from Google?
The truth is, Google is among the few companies that built one of the most complete AI ecosystems in the world and most people don’t even realize how deep, interconnected, and powerful it has become. 

From advanced multimodal models to coding agents, business assistants, creative tools, and enterprise AI platforms, Google’s ecosystem covers nearly every category of modern AI development.

Here’s a breakdown and what each part does:

🔸The Models
Google’s core family of Gemini and Gemma models power reasoning, multimodal understanding, fast inference, and lightweight on-device AI.

🔸Build & Code
A full suite of tools which include App Builder, AI Studio, Code Assist, Antigravity, and Jules. They help developers prototype apps, debug code, automate tasks, and build production-ready agents.

🔸Video & Motion
Tools like Veo, Lumiere, Flow, and VideoFX generate cinematic footage, consistent characters, smooth motion, and AI-powered video storytelling.

🔸Image Creation
Imagen 3, Nano Banana, Google ImageFX, Whisk, and Stitch create photorealistic images, editable UI layouts, artistic styles, and precise controlled-generation outputs.

🔸Assistants & Business
Gemini Live, Gemini Gems, Workspace Gemini, Google PomellI, and NotebookLM act as reasoning assistants, brand agents, workflow automators, and personalized long-term AI personas.

🔸Additional Ecosystem Pieces
Vertex AI, Gemini for Firebase, Gemini Nano, Creative tools, and AI Safety systems bring enterprise orchestration, mobile integrations, on-device intelligence, and responsible AI safeguards.

However, this doesn’t mean this is the only choice you have when it comes to building state of the art AI systems. Would love to know which one you’re a big fan of. 
#AI

The fastest way to get ahead in AI? 
Build the skills everyone will need in the next 12 months.

Mastering LLMs isn’t about knowing prompts, it’s about understanding the entire ecosystem behind the model.

If you can learn these 14 skills, you won’t just use AI — you’ll engineer it.

1. Understanding the LLM Ecosystem
Grasp how models, context windows, embeddings, RAG, prompts, and vector DBs all fit together so you can design end-to-end AI systems confidently.

2. Adoption Challenges & Risks
Learn the technical, operational, and ethical risks of real-world AI deployment, from hallucinations to prompt brittleness to evaluation gaps.

3. Evolution of Embeddings
Understand how text is represented mathematically, from TF-IDF to dense vectors, and choose the right embedding approach for real NLP tasks.

4. Attention Mechanism & Transformers
Master how transformer models process context using self-attention so you can reason about model behavior and limitations.

5. Designing Retrieval with Vector Databases
Learn vector search, indexing, hybrid retrieval, reranking, and how vector DBs power scalable RAG applications.

6. Semantic Search
Move beyond keyword search and use embeddings to retrieve meaning-based results that match user intent.

7. Prompt Engineering
Design structured, repeatable prompts using CoT, ReAct, few-shot, multi-modal prompting, and learn how to avoid vulnerabilities like injection.

8. LLM Fine-Tuning
Understand when fine-tuning is actually needed and learn methods like SFT, DPO/RLHF, LoRA, and QLoRA to adapt models safely.

9. Orchestration with LangChain
Build scalable LLM apps using document loaders, chains, agents, memory, output parsers, and retrieval pipelines.

10. Retrieval-Augmented Generation (RAG)
Combine real-world data with LLMs to reduce hallucinations and support enterprise-grade search and knowledge workflows.

11. Evaluation & Monitoring
Learn how to measure LLM accuracy, safety, behavior drift, and reliability - a critical skill for production AI.

12. Model Deployment & Scaling
Ship LLM apps with APIs, memory management, batching, caching, versioning, and cost-optimization strategies.

13. Agents & Autonomous Workflows
Use agent frameworks to let LLMs plan, decide, call tools, run sequences, and automate multi-step operations.

14. Data Engineering for LLMs
Prepare clean, structured data pipelines so LLMs have high-quality inputs, the foundation of every successful AI system.

LLMs aren’t mastered by learning prompts alone, they’re mastered by understanding the full stack: embeddings, retrieval, orchestration, fine-tuning, and evaluation. Build these skills and you’ll be ready for any AI role in 2026.

14 key skills to master LLM

Check out this simple breakdown of the four core LLM architectures shaping most of today’s advanced systems.

Not all LLMs reason the same way. Building reliable AI systems, or choosing the right model for a workflow requires a clear mental model of the architectures that drive them.

Let’s review:

1.🔸Decoder Only (GPT, LLaMA, Grok)
Generates text by predicting one token at a time using previous context.

Best for generative workloads, conversational systems, and multi step reasoning.

2.🔸Encoder Only (BERT, RoBERTa)
Processes text with full bidirectional context to learn dense semantic representations.

Best for classification tasks, retrieval, and embedding generation.

3.🔸Encoder Decoder (T5, FLAN T5, BART)
Separates understanding from generation. The encoder builds meaning, the decoder transforms it.

Best for translation, summarization, and any structured text to text transformation.

4.🔸Mixture of Experts (Mixtral, Gemini 1.5)
Routes each input to a subset of specialized expert networks.

Best for scaling model capacity while controlling compute cost.

These architectures go beyond academic details to determine latency profiles, context handling, fine tuning strategies, and how models behave under real workloads. A clear grasp of them helps you design better pipelines, choose the right foundation models, and understand why two LLMs can produce very different outcomes on the same task.

Let me know your thoughts!

#LLM

I believe the next wave of AI systems will not be defined by bigger models but by smarter orchestration. Salman Paracha got my attention on Arch Router, which introduces a routing approach that aligns model selection with human preferences instead of benchmark scores. You should absolutely check it out, as I know machine alignment is one of the most talked about challenges at enterprise. 

Arch Router treats routing as a semantic matching problem between a user query and a set of natural language policies built from a Domain and Action taxonomy. This design gives teams direct control over how queries should be handled and removes the need to retrain routers whenever new models are added. 

The authors trained a compact 1.5B model that outperforms top proprietary models in multi-turn routing while running at a fraction of their latency. The result is a practical blueprint for multi-model systems that are fast, interpretable, and aligned with real usage rather than synthetic metrics.  

Let me know your thoughts on this one. 

Salman and I will be sharing more on the subject.

#LLM

Arch Router: Aligning LLM Routing with Human Preferences

Super rare having the privilege to dine with world-class AI leaders from Amazon, Zapier, Total Neural Enterprises (tne.ai), Google, Salesforce, and Braintrust, while nerding out about everything, especially the exciting future ahead!

I enjoyed learning so much from all these giants, which include Rich Tong, chee chew, Amit Gupta, Jenny Jau, 💡Prashant Singh A., Ameya Bhatawdekar. Check out their profiles, you’ll be amazed! Ken Morimoto made this possible, kudos to you brother! 

We chatted about the future of AI Agents and creative frameworks to make them more reliable. We had an in-depth discussion about the challenges executives face which slow AI adoption (Rich, I can’t wait to get the ball rolling with you). 

We hashed it out on Quantum Computing, Edge AI, Autonomous Robots, AI Creativity (Will we ever have an Newton-like AI who can create something out of nothing?). AI Creativity took us to the ARC AGI Leaderboard as a measure for autonomous creativity. 

We nerded out - yes, nerded - about theater (Issaquah’s Village Theatre is a good place to catch good shows).

Lastly, we shared wisdom about some of the books that shaped our mental model of the world and changed our lives. Most books on the list were not about AI. I admire the human dimension!

Thank you all for a wonderful chat. Despite the busy life, you gave Ken and I some of your precious time, and we enjoyed being a sponge. 

Ken is in NY this week for more exec dinners, do reach out to him if you’re in the area. Space limited.

What a great time we live in!

BOOKS:

• Life 3.0, Max Tegmark
Physicist and MIT professor exploring artificial intelligence, consciousness, and the long-term future of humanity.

• Determined, Robert M. Sapolsky
Neuroscientist and biologist examining free will through the lens of biology, neuroscience, and behavior.

• A New Earth, Eckhart Tolle
A spiritual exploration of ego, consciousness, and personal transformation.

• In the Middle, Nancie Atwell
A seminal work on writing, reading, and teaching, centered on student-driven learning.

• Team of Teams, General Stanley McChrystal
Leadership lessons on adaptability, decentralization, and decision-making from modern military operations.

• Hide a Dagger Behind a Smile, Kaihan Krippendorff
Ancient Chinese strategic thinking applied to modern business and competitive environments.

• Bhagavad Gita, Vyasa
A foundational philosophical and spiritual dialogue from the Mahabharata, centered on duty, action, and self-realization.

• Thinking in Bets, Annie Duke
A practical framework for decision-making under uncertainty, informed by professional poker and probability theory.

• The Holographic Universe, Michael Talbot
An exploration of reality through quantum physics, neuroscience, and mysticism.

• Sea People, Christina Thompson
A historical and cultural account of the Polynesian navigators who explored and settled the Pacific Ocean.

#AIAgents
#AICreativity
#QuantumComputing
#AutonomousRobots

No alternative text description for this image

🚀Is multi-agent better? Not necessarily! Google DeepMind just released one of the most important papers of 2025 when it comes to deploying at-scale agentic systems.  

“Towards a Science of Scaling Agent Systems” introduces a quantitative framework for understanding when single agent systems outperform multi-agent systems and when coordination actually helps. 

Turns out agent performance is driven by task structure and coordination cost, not agent count. In controlled experiments across finance, web navigation, planning, and workplace tasks, the researchers show that multi-agent systems can improve performance by over 80% on parallelizable tasks like financial analysis, yet degrade performance by up to 70% on sequential tasks where coordination fragments reasoning.

There are three dominant forces identified: 

🔹A tool coordination tradeoff where tool-heavy tasks suffer from multi-agent overhead under fixed token budgets. 

🔹A capability ceiling where once a single agent crosses roughly 45 percent accuracy, adding coordination produces diminishing or negative returns. 

🔹An architecture-dependent error amplification, where independent agents propagate errors aggressively while centralized orchestration contains failures through validation bottlenecks.

What matters most is alignment. Centralized coordination works well for decomposable reasoning. Decentralized coordination helps in exploratory environments like web navigation. For strictly sequential planning, single agents remain superior. 

To prove this, the researchers built a predictive model that explained over half of observed performance variance in the paper, and correctly selects the optimal architecture for most unseen tasks.

Afterall, scaling agents is a systems design problem and far from being a prompting trick. The developers that will stand out by building reliable agentic systems will understand how to measure coordination efficiency, overhead, and task decomposability, before adding complexity.

Do check this paper out!

#AIAgents

taplio

The Battle

George Firican 💡

Greg Coquillo

Followers

Posts

Engagements

Likes

Performance ScoreGeorge Firican 💡's LinkedIn Performance

Performance ScoreGreg Coquillo's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

George Firican 💡

Greg Coquillo

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}George Firican 💡's LinkedIn Performance

Performance ScoreGreg Coquillo's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreGeorge Firican 💡's LinkedIn Performance