OpenAI and Google: Rivalry in Enterprise AI

OpenAI Launches Massive AI Deployment Initiative

OpenAI recently announced an ambitious project aimed at creating a new company with an initial investment of over $4 billion. This initiative is designed to help organizations build and deploy artificial intelligence systems. As part of this strategy, OpenAI has also decided to acquire Tomoro, an AI consulting firm, to accelerate the development of this new unit.

Since its early models captured public attention, OpenAI has been striving to secure contracts with businesses and establish a strong presence in the corporate world. The goal is to deploy its AI on a large scale. This project, which will primarily be under OpenAI's control, comes at a time when its competitor Anthropic is experiencing notable success with its Claude models, which have been quickly adopted by companies. The new entity, named OpenAI Deployment Company, aims to integrate engineers specialized in deploying cutting-edge AI within organizations. These engineers will work closely with various teams to identify areas where AI can have the most impact, according to OpenAI.

The acquisition of Tomoro, a consulting firm that helps businesses integrate AI, will bring approximately 150 experienced AI engineers and "deployment specialists" to the new unit from day one. Founded in 2023 in partnership with OpenAI, Tomoro counts among its clients companies such as Mattel, Red Bull, Tesco, and Virgin Atlantic, according to its website.

Google Strengthens Its AI Engineering Team

According to The Information, Google plans to hire hundreds of engineers to help its clients adopt its enterprise-focused AI products. These new "frontline deployed engineers" will form a new team within Google Cloud, as indicated by Thomas Kurian, the head of the unit, on LinkedIn. Matt Renner, Google Cloud's revenue director, added in a separate post that this initiative would allow Google to present itself to its clients with more technical resources, rather than a large number of salespeople.

This announcement is part of a series of recent initiatives in the sector, as tech companies mobilize human teams, often referred to as "frontline deployed engineers," and establish partnerships with consulting firms to encourage the use of AI-based technologies aimed at automating work. Recently, OpenAI launched the OpenAI Deployment Company in partnership with consulting and investment firms. The week prior, Anthropic announced the creation of a joint venture with private equity firms to sell its AI to the clients of those firms.

AI and Businesses: A Historical Perspective

It is tempting to make sarcastic remarks about the fact that general artificial intelligence (AGI) does not yet seem ready to deploy AI, but I prefer to say "as expected." In the article "Business Philosophy and the First Wave of AI" from 2024, I argued that the appropriate analogy for AI in business was not SaaS, but rather the first wave of computing in the 1970s.

AI agents are not mere copilots; they are replacing humans. They perform tasks in place of humans, such as in call centers, and benefit from all the advantages of software: constant availability and scalability according to demand. Benioff does not talk about improving employee productivity, but rather that of businesses; the term used for employees is "augmented," which sounds better than "replaced." The ultimate goal is to improve business outcomes. This technological philosophy aims to enhance the bottom line of large companies.

This framework fits well with the wave of mainframes: accounting and ERP software made businesses more productive and generated positive business results. "Augmented" employees were managers receiving more accurate and faster reports, while those performing that work were replaced. The decision to proceed with this change did not depend on frontline employees, but on leaders who decided to embark on it.

I do not think the Deployment Company is there to help employees use chatbots; it is even more evident with the private equity firms with which OpenAI and Anthropic are making deals. I expect an increasing number of agreements where private equity firms buy software companies with reliable cash flows and proceed with significant layoffs, forcing AI to compensate while addressing stock-based compensation issues.

I do not know if the mandate of the Deployment Company will be as severe, but I assume it is a company engaged by management to fundamentally rethink business processes in a way that has not been done since the era of mainframes.

The Crucial Role of Data

This raises the question of data, and while Benioff boasts about all the data Salesforce possesses, he does not have it all, and what he has is scattered across the multitude of applications and storage layers that make up the Salesforce platform. Indeed, Microsoft faces the same problem: although their vision for Copilot includes APIs for third-party "agents" — in this case, data from other companies — the reality is that an effective agent — that is, a worker replacement — must have access to everything in a way that allows it to reason. The ability of large language models to handle unstructured data is revolutionary, but the fact remains that better data leads to better outcomes; explicit and reasoned data, for example, is integral to the functioning of o1. In this regard, the company that intrigues me the most, for what I believe to be the first wave of AI, is Palantir.

This integration resembles the illustration from the company's webpage for Foundry, what they call "The ontology-powered operating system for the modern enterprise":

What is notable in this illustration is how deeply Palantir must get involved in a company's operations to achieve its goals. This is not a consumer SaaS application that your team leader pays for with their credit card; it is SOFTWARE of the type that Salesforce sought to surpass.

Comparisons and Reflections on Palantir

Kurian from Google, by the way, dismissed any comparison with Palantir during a Stratechery interview last month:

All of this makes perfect sense, especially this part regarding the Knowledge Catalog, which aligns with my thinking. I wrote a few years ago about the importance of this entire layer and the need to understand it; it is somewhat of a significant challenge to set this up. You have a sort of analogy, let's say, with a Palantir setting up their ontology. They have FDEs on-site, multi-month projects for that. You have OpenAI talking about Frontier, their layer of agents, and they are partnering with all the tech consulting firms to develop this. Will it require a lot of on-the-ground presence for this graph to function and be operational in a way that allows your agents to operate effectively across it?

TK: We are not competing with Palantir; we are not building a semantic dictionary or ontology. What we are doing is that today, I will give you the closest analogy.

TK: Today, when you use a model, let’s say you are using Gemini, and you ask a question, Gemini goes through reasoning, and then it shows you a citation. A citation is: "How did I answer the question and what is the source I drew from?"

Now imagine that this citation is a query that needs to go into a folder in, for example, a storage system, because there are documents there, and a database because, for example, in a part number, just think that there is a part number document that lists all part numbers and is located on a disk, and this part number that you need to extract to say that it is the modem that the person is coming to repair, and that is mapped to a table in a database.

So what the graph does, we use Gemini, so we do not need humans, we use Gemini to say: "Hey, go read all these documents in these disks and extract the information from them, then match that to the database table that has the reference to the part number," and so when Gemini turns around and says: "I have this query on how much modem inventory they have," the first thing it does is it says: "Okay, go to the Knowledge Catalog and it says that the modem is part number one, two, three, four, five," then it says: "By the way, the table in the database that has the inventory information on this part number is this table, here is an SQL," this improves the quality of what we generate and when it answers the question, it shows again — back to your, "Trust my data," it shows a foundational citation saying: "This is where we got it from."