A Note from Hamza & Angela: The 2025 Roller Coaster & Diving into AI-enabled Services

Whew! What a week it’s been!! Feels like 2025 is going to be a roller coaster and even crazier than 2024 when it comes to AI and startups. Even the aliens watching us from afar waiting to invade (cue the Fermi Paradox), will be scared of the exponential progress of tech on earth this week! 👽

So much to catch you up on! SOO, seems like people are loving this insights section, so I’m going to continue and build on the previous piece (where we covered Verticalized AI Agents), to discuss why we’re excited about a more hidden area - AI Enabled Services, too. But first, a quick mention on what happened this week!

This Week in AI (One Year in Normal Time)

I won’t go too deep as you’ll have read the lowdown all over Linkedin, X and other newsletters anyway, but what the heck!

From the Project Stargate investment ($500B into American AI infra), to Deepseek’s R1 causing a watershed moment, to OpenAI releasing it’s UI-based agent “Operator” (equivalent to Anthropic’s Computer Use). All of this is going to disrupt knowledge work, industry and the world generally. Strap in for a crazy year ahead.

I can’t move on without giving a shoutout to Deepseek and what R1 means for startups/founders. This looks like another before and after moment in AI, similar to the launch of ChatGPT on November 30, 2022.

1️⃣ Firstly, open source is now on par with closed models and privatized AI (sorry to all the OpenAI and Anthropic investors in the room). Meta’s Head of AI, Yann Le Cun, says it’s less about US vs China, and more about the fact that “Open source models are surpassing proprietary ones”. Marc Andreessen of a16z called R1 one of the most impressive tech breakthroughs he has seen and a “profound gift to the world”. The beauty is that the whole world can profit from this open work, not just a few entities.

2️⃣ Founders building applications should be clapping right now as this will force all model providers to release the latest models faster and cheaper - the foundation model layer being hyper competitive is great news. Furthermore, the commoditization of intelligence shifts value at the model layer to other kinds of capabilities like agentic behaviour, creating more innovation in new areas.

3️⃣ R1 proved that models can be created as smaller models and be on par with larger models (R1 even beats OpenAI’s o-1 model on some benchmarks like math problems), for a fraction of the cost. And the cost of running the model itself is approximately 95% cheaper for a performance comparable to OpenAI’s o-1. This is great news for startups who are paying OpenAI and Anthropic exorbitant token fees every month to run models within their companies (we’ve heard some startups are paying in the millions per month for running on the larger models).

4️⃣ Not to mention the unique research and engineering feats of R1 which takes reasoning to another level. The Deepseek team used reinforcement learning (a reward mechanism where AI is rewarded for getting the right outcome), so the model learns by trial and error, getting really good at thinking logically. There’s also an element where the training algo no longer needs human data for training but can test the AI model for specific types of problems that are verifiable - distorting the thinking that human data is the limit. These training methods are all open source so the AI community can benefit and build better reasoning models faster.

5️⃣ The questions is, will the Chinese government steal our data when using Deepseek’s model? Potentially, if using Deepseek’s official app. But the model weights are fully open source so you can run them on device locally and keep your data private. This is great for early stage startups and small businesses who want to move quickly, but we expect enterprises are likely to stay away from this model given the Chinese affiliations.

The Hidden Value in AI Enabled Services zzzzzzzzzz…Okay, I know it SOUNDS boring, but hear me out about this $4 trillion opportunity

What do we mean by services being enabled by AI? Many services jobs (legal, customer support, etc.) today involve repetitive tasks, making them ideal candidates for AI-driven automation. Agentic AI is capable of decision-making that can alter how a job is executed, taking a bite out of the services market as a whole whereas traditional software has simply helped increase productivity in a job. But why is this so interesting? Consider the sheer scale of these markets:

For scale, the global Saas market is worth $400 billion, whilst the services market is worth $4 trillion. The traditional viewpoint with services or agency businesses is that they are not scalable and so their margins are not as high as Saas companies. But by making these AI enabled, with humans still a strong element of the service, you can increase profit margins to the point where it can be scalable and very profitable indeed.

Law as an example

Let’s take law as an example. Given my background in law and stint as a legal tech product manager in a past life, I (Hamza) spent time working with an AI legal tech company that was serving financial services and asset management clients. Asset management firms go through thousands of NDAs and high volume routine contracting yearly. Paying fees to law firms to review those contracts can be extremely costly. That’s when players like Ontra came into the market offering offshore NDA reviewing services (they had freelance lawyers reviewing the NDAs at a fraction of the price).

Then came legal tech startups going after that same market, but pitching even lower costs due to supposedly higher automation. However, what they were really doing (and still are) is having human lawyers do the work in the back and pretending to clients that it’s an AI bot in the front. The beauty is that they were making revenue from selling services to asset managers, whilst also building out the technology with the engineering function to automate the service which of course takes time to develop. I don’t know about you, but I think this is a pretty cool alternative to the high capex model of building tech companies with little or no revenue for the first few years that most Saas companies go through. Plus, you get direct insights about how and what to automate through the services arm.

Now, couple this with developments in foundation models and AI agents (enhanced reasoning, UI based task execution and more). You can automate a lot of those workflows, have a few humans there to deal with client relationships and managing the AI agents, whilst maintaining the same or an even larger client base. Thus moving the needle of service level profit margins up much closer to that of Saas level profit margins.

The AI Rollup Opportunity

Where this gets even more interesting, is where this can lead to roll up opportunities (where an entity buys up companies in a certain niche to create a bigger player in a market, usually leading up to an IPO) similar to private equity roll ups, but driven by an AI startup. Getting your AI agents to the level where they can carry out the workflows, acquiring other services and agency businesses within your niche, accessing their client bases, introducing these AI agents to move to stronger profit margins and…I’ll leave the rest to your imagination…

What this takes is services firms and agencies taking an AI native approach and mindset from the get-go.

There are other service markets here too such as accounting, consulting, etc. What other high TAM services markets can you think of? And man, those TAMS are big.

From RAGs to SLMs with Big Poppa?

P.s. On a side note, we loved this podcast with George Sivulka, founder and CEO of Hebbia, where he said that RAG (retrieval augmented generation) doesn’t really work too well on real world examples. I also spoke to a founder last week who mentioned that Small Language Models discounted the need for RAG when it comes to accuracy. The latter was an AI Hot 100 podcast we’ll be releasing soon, so more to come.

Until next time, yours truly,

Hamza Zaveri & Angela Mascarenas

Any questions or feedback on this section? Reach out to us [email protected]

📈 Venture Deals To Watch Out For

• Kaya AI, an AI construction supply chain intelligence platform, raised $5.3M in its Pre-Seed funding. (link)

• Orchid Security, an identity protection service company, raised $36 million in its Seed funding. (link)

• Rembrand, an AI product placement and content company, raised $23M in its Series A funding. (link)

• NEURA Robotics, a cognitive robotics company, raised €120 million in its Series B funding. (link)

• Komainu, a digital asset service provider and custodian, raised $75 million in its Series B funding. (link)

• ElevenLabs, an AI company that creates and edits synthetic voices, raised $250 million in its Series C funding. (link)

• Synthesia, an AI media generation company , raised $180 million in its Series D funding. (link)

• Clay, a data enrichment and automation company, raised $40M in its pre-emptive Series B funding. (link)

⚙️ New Tools & Product Launches You Won’t Want to Miss

• Napkin AI - an AI that generates graphs and more from text for business storytelling. (Napkin AI)

• Fyxer AI - an AI assistant that organize emails, take notes and more. (Fyxer AI)

• Oxylabs - a tool for web data gathering at scale and proxy solutions. (Oxylabs)

• PixVerse - a multi-modal AI-powered video creation platform. (PixVerse)

• Lambda - an AI and ML infrastructure provider. (Lambda)

• Hunyuan 3D from Tencent - Tencent’s AI-powered engine for 3D models. (Hunyuan)

• DeepSeek - well well well, who isn’t talking about it right now? (DeepSeek)

🎞️ Highlights from The AI Hot 100 Conference: Investing at the Frontier - The Hidden Value in AI Panel

Hear from Marc Andrusko (Investing Partner, a16z), Nihal Mehta (Founding General Partner, Eniac Ventures), Lisa Dolan (Managing Director, Link Ventures) and moderated by Hamza Zaveri about Investing at the Frontier: The Hidden Value in AI (from The AI Hot 100 Conference in NYC, September 2024).

Top Jobs in AI

Join a high growth funded ($5M+) startup, having <100 employees, which are actively HIRING.

Speak ($78M Series C) - AI enabled language tutor

• ML Engineer (Link)

• AI Product Engineer (Link)

• Head of Talent Acquisition (Link)

• US GTM Lead (Link)

Genesy ($5.2M Series A) - AI-driven B2B lead generation and outreach automation platform

• Frontend Developer (Link)

• Product Designer (Link)

• Marketing Content Specialist (Link)

Kamiwaza AI ($8.6M Series A) - GenAI orchestration engine enabling applications to run on top while keeping enterprise data private and secure

• GenAI Dev Advocate (Link)

• GenAI Platfrom Engineer (Link)

• Technical Chief of Staff (Link)

Prepared ($27M Series B) - AI assistance for every stage of emergency response process

• Engineering Manager (Link)

• Sr. Product Designer (Link)

• Information Security Engineer (Link)

MindGard ($8.2M Series A) - Detect and remediate against attack vulnerabilities in AI models

• AI Security Analyst (Link)

• Software Engineer (Link)

• Account Executive (Link)

Looking to hire in AI? Apply here to add your roles to our Job Board.

Interesting Posts

Deepseek + Claude: Use the new Deepseek reasoner through Anthropic’s Claude desktop, extracting the thinking portion of Deepseek with a Clause interface. See it here.

Is AI making students dumber or lazier? A study explored how tools like ChatGPT affect learning motivation, processes, and outcomes. Researchers worked with 112 students divided into four AI vs non AI groups. The findings show that ChatGPT can help students perform well in structured tasks, but it might also make them overly dependent on AI as they often skipped critical thinking steps, reducing their ability to think critically or self-regulate their learning. For AI to truly enhance learning, there’s a need for proper guidance and support to ensure students stay engaged and develop their thinking abilities over time. Read the study here.

Can AI Agents reflect and correct their actions? Agent-R is an iterative self-training framework that allows language agents to reflect and correct their actions in interactive environments. See here.

Upcoming Events 📆

📢 Interested in sponsoring? Reach out to us at [email protected] to get your brand in front of thousands of AI entrepreneurs, researchers, investors, executives, and AI leaders.