Don’t Miss This: DeepSeek R1 Insights + 2 Free Resources | AI:Unlocks™️ Newlsetter 1/27/25
There’s so much to unpack from this past week, and I’m excited to cover it all for you today.
Before we dive in, let me briefly introduce this newsletter.
What AI:Unlocks Is Not:
A news roundup.
An educational resource.
A stream of pontifications or speculative insights about the future of AI.
It’s not only focused on ethics, human-centered AI, or success stories of AI adoption in business.
What AI:Unlocks Is:
It’s all of these things!
...Tailored to what you, the business leader, care about and need to know.
Let’s get started.
Why I’m Covering DeepSeek Now
When DeepSeek R1 dropped, the swirl was swirling hard. The headlines made it seem like OpenAI and Anthropic were about to be rendered obsolete overnight. I waited…
I wanted to let the dust settle, watch how developers reacted, and bring you the most relevant takeaways—not just in-the-moment excitement, but the real business implications after the technical community had weighed in.
Here's what I see:
DeepSeek R1 is a technical breakthrough, but it’s not an enterprise solution yet.
It lacks the robust infrastructure, API stability, and multi-model orchestration that OpenAI and Anthropic offer.
In real-world enterprise applications, AI isn’t running in a vacuum—it’s making hundreds of API calls, working across multiple specialized models, and integrating into existing workflows.
That’s where DeepSeek falls short (for now).
But let’s not mistake a temporary gap for a permanent one. More on this in a moment. First,
Let's Get Up To Speed
DeepSeek, a Chinese-built AI model, has been making waves these past weeks. What is DeepSeek R1?
It’s a model trained entirely through reinforcement learning no pre-training, just iterative learning from feedback.
Why is that significant? Because it marks a shift in how AI systems learn. Instead of relying purely on massive datasets and supervised training, DeepSeek R1 refines itself through direct experience. It’s closer to how humans (or animals) learn—trial and error, iterative refinement.
Why This Matters to You
DeepSeek’s innovations have fundamentally reshaped the AI landscape, making high-performance AI more accessible, efficient, and cost-effective.
Here’s why you should care:
It’s open-source.
Its reasoning capabilities rival OpenAI’s top-tier models.
Its API usage is cheaper—substantially so.
Cheaper. Better. More accessible, once it's enterprise ready.
But that’s not the only shift DeepSeek represents:
FOBO: The Fear of Being Outpaced
DeepSeek’s emergence highlights something every executive is grappling with.
How do we invest in AI while staying flexible in an environment that changes weekly?
100% of executives I talk to articulate some version of this conundrum.
The answer? A structured, pragmatic pilot-to-scale approach.
Develop an AI-First Mindset
Encourage your team to fully utilize AI tools through structured, operationalized workflows—not just experimental prompting.
Build a Strong Data Practice
DeepSeek’s performance proves that learning velocity matters more than raw data volume. Ensure your data isn’t just big—it’s fast, well-labeled, and actionable.
Assign Ownership
AI implementation isn’t plug-and-play. Identify a strong executive sponsor and clear accountability structures.
Tip: Pilot in low-risk, high-impact areas before rolling out org-wide changes. Avoid tangled spaghetti transformations.
Tip: Target 70% AI tool usage in daily workflows to unlock measurable efficiency gains, revenue lift, and avoided costs.
Where The Value Is Shifting
Open-source models are evolving at warp speed. What DeepSeek lacks today in deployment sophistication, the open-source ecosystem will build out soon enough. And when that happens, AI economics will shift again.
The biggest macro-trend DeepSeek introduces is the market bifurcation between model providers and application innovators.
--> As models become commodities, it will be more difficult for them to compete. Yes, important for pushing the boundaries of what models can do.
But,
--> The others will be those who can take these foundational models and integrate them into compelling, user-centric products and services (and those enterprises that can integrate them toward their own strategic advantages) .
The Inside Baseball – What Are Developers Saying?
The reaction in the developer community has been mixed, polarizing, even.
Excitement – Many are thrilled about training models purely through reinforcement learning, which was previously considered inefficient.
Caution – Some worry about interpretability: If a model thinks in ways beyond human comprehension, do we restrict it to keep it readable, or let it evolve freely?
One developer on LinkedIn put it well:
“The smarter these models get, the less manageable they are.”
Which leads to an even bigger question.
What Actually Makes DeepSeek Different?
The technical shift is more than just reinforcement learning.
DeepSeek’s architecture introduces
Mixture of Experts (MoE): A modular structure that activates only relevant parts of the model, dramatically reducing compute costs.
Multi-Head Latent Attention (MLA): A breakthrough in memory efficiency, slashing memory overhead by 93.3%, allowing for longer, cheaper interactions.
Distillation: Smaller, highly capable models that democratize access to advanced AI—without enterprise-level costs.
}}}
AI is no longer just a "who has the most compute" game. Competitive advantages are shifting to who can deploy, optimize, and scale AI the smartest.
Where AI’s Competitive Edge Is Moving Next
As models become commoditized, the real competitive advantage won’t come from having the best model—it will come from workforce orchestration.
Who can make these models work for their business the best?
That’s the next battleground.
This is where you’ll see the rise of Platform AI—integrated systems that seamlessly manage multi-model workflows, dynamic AI routing, and enterprise-grade orchestration.
The race isn't about who has the next model, it's about who masters workforce orchestration.
Knowing when to call which model for different tasks.
Balancing cost vs. performance dynamically.
Seamlessly routing AI outputs into automated decision-making systems.
DeepSeek isn’t there yet, but it signals a future where open-source models will be. And when they are, the organizations that have mastered AI orchestration—not just AI procurement—will win.
Final Thoughts + Resources
Let’s bring this back down to earth, to what's important now - execution.
For those who haven’t downloaded it yet, here are two resources to get started today:
Your AI Study Guide
A structured roadmap for leaders and teams to accelerate AI adoption effectively.
The Spreadsheet Method for Brand-Aligned Content
A smart way to prompt GenAI for consistent, on-brand content creation—without the usual chaos.
That’s all for now.
As we move into an AI-first future, remember: Who you are is your greatest asset.
Have a great week,
Megan
p.s. I'll be speaking at BI Worldwide this month - so far 130 executives from across functions at the Twin Cities' largest enterprises will be there. If you're near Minneapolis, we'd love to see you.
https://info.biworldwide.com/mpls-ai-enablement
p.s.s - Of course I had to remark on ChatGpt's SuperBowl commercial. I'll say this: Human creativity is irreplaceable, and we saw that here. While I wiped tears from my eyes watching Google's gemini commercial, ChatGPT wholly missed the mark - the emotional impact fell leagues short of the impact its had on society the past 2 years.
