Putting AI to work: The latest from MIT Sloan Management Review

New insights from MIT Sloan Management Review classify the main types of artificial intelligence startups and how companies can benefit from working with them; provide tips for using agentic AI coding tools for knowledge work; and explore why artificial intelligence isn’t ready to contribute to large-scale productivity improvements.

Get to know the 6 types of AI startups

Venture capital firms are insisting that startups in their portfolios incorporate generative AI into their product offerings or internal operations. There’s a good reason for that, according to MIT Initiative on the Digital Economy fellow Thomas Davenport and Babson College professor Jeffrey Shay: Startups don’t face the organizational or technical barriers that can stall enterprise AI implementations.

The challenge, though, is knowing where, when, and how to use AI, as well as understanding who’s developing AI tools. 

Davenport and Shay classify AI startups into six types to provide clarity to would-be adopters: 

1. Originators build commercially deployed foundational models.
2. Explorers look at future use cases, such as agentic AI and quantum AI.
3. Infrastructure builders create the data, application programming interfaces, and frameworks needed to use generative AI.
4. Enhancers apply general-purpose AI models to specific industries or problems.
5. Optimizers use AI to transform how they operate.
6. Experimenters test AI without a budget, road map, or integration; this is by far the largest cohort of companies.

Potential corporate customers should ensure that they work with companies that match their needs and expectations. Organizations with existing AI capabilities may want to turn to infrastructure builders to accelerate development, for example, while those looking for quick wins for specific functions should seek enhancers as partners.

Read ‘Six types of AI startups, explained’

Don’t assume AI will improve productivity

Throughout history, the impact of technology has not been “preordained,” MIT Institute Professor and Nobel laureate Daron Acemoglu argued in a podcast interview in which he expanded on ideas presented in his 2023 book, “Power and Progress.” “We have a lot of agency, a lot of choice in shaping the future of technology, and different futures correspond to different winners and losers,” he said.

AI is no exception. It is versatile enough to provide different versions of the future depending on how it’s used — each with different outcomes for economies, industries, and workers. 

But right now, Acemoglu argued, organizations are choosing to use AI as automation technology when, in reality, it is a form of information technology. This explains why AI isn’t improving productivity at a macroeconomic level: Automation benefits capital owners. Workers, meanwhile, would benefit from the ability to gain insights and make judgments from the information that AI systems aggregate.

It doesn’t help that existing architectural and economic models make it difficult to build AI tools this way. Theoretically, electricians could use AI tools to understand why equipment isn’t working, or nurses could use it to generate recommendations for treating a patient. For that to happen, though, AI tools need to be optimized for the task, reliable, and trained on domain-specific information. All of that takes time and capital to accomplish.

What needs to change? Acemoglu believes organizations must determine where automation can address bottlenecks without overdoing it. He suggested the creation of “pro-worker, pro-human technologies” that decentralize access to information and support front-line decision makers. This would allow the gains from “technological betterment” to be shared more equitably, as was the case when innovation led to widespread wage growth after World War II. 

But many of today’s innovators — including startups — have chosen to align with capital owners in the pursuit of acquisition and wealth creation. “Perhaps we should really be much more vigilant in mergers and acquisitions,” Acemoglu said. “That could lead to very different dynamics.”

Listen to ‘AI is not improving productivity’

Leading the AI-Driven Organization

In person at MIT Sloan

Register Now

Use agentic AI coding tools for knowledge work

You might think that agentic AI coding tools are only for software developers, but MIT Sloan professor of the practice Rama Ramakrishnan writes that tools like Claude Code can also be used for knowledge work. The three technical building blocks of agentic AI coding tools help explain why:

• Multistep reasoning breaks a task into a sequence of steps and creates an action plan for executing them.
• Adaptive execution observes the results of each step and course-corrects if something doesn’t look right.
• Tool use involves processes executed by an external system, whether it’s connecting to a database or running a command.

These technical building blocks indicate that agentic AI coding tools can not only read files but also remember what’s in them, re-execute an analysis when there are new source materials, and run multiple tasks at once. As a result, Ramakrishnan writes, the tools are well suited for tasks such as conducting competitive intelligence, preparing for meetings, and creating different versions of marketing campaigns. (Read the article for more detailed guidance on writing and refining prompts.)

Ramakrishnan offers some words of caution for executives before they get to work. One is to ensure that coding tools have access only to trusted folders and files. Excluding documents from outside sources reduces security risks known as prompt injections, which manipulate an AI tool into unauthorized action. And users should review a tool’s proposed actions before approving them, because no agentic AI coding tool is 100% accurate, and mistakes can happen.

Read ‘AI coding tools for knowledge work: What executives need to know’

This article draws on insights from MIT Sloan Management Review, which leads the discourse about advances in management practice among influential thought leaders in business and academia. The publication equips its readers with evidence-based insights and guidance to innovate, operate, lead, and create value in a world being transformed by technology and large-scale societal and environmental forces.