Eight examples of artificial intelligence in action

Until recently, many companies have lived in a sort of purgatory of artificial intelligence (AI) development, conducting endless pilots and proofs of concept, but bringing very few AI-enabled projects through to enterprise production.

That’s changing fast. According to the latest McKinsey Global Survey on AI, conducted in July 2024, 78% of respondents say their organizations use AI in at least one business function, up from 72% in early 2024 and 55% a year earlier. The use of generative AI (GenAI) has also risen, with 71% of respondents telling McKinsey that their organizations regularly use GenAI in at least one business function, up from 65% in early 2024. And AI investments continue to grow. More than three-quarters (78%) of business leaders said their organization plans to increase its overall AI spending in the next fiscal year, according to the Deloitte AI Institute’s 2024 State of Generative AI in the Enterprise report.

How much business value AI is delivering thus far, however, may be a matter of debate. For example, 21% of C-suite survey respondents reported they feel GenAI is already transforming their organization, compared to only 8% of non-C-suite respondents, according to the Deloitte report. And that’s just one type of AI, a catch-all phrase for cognition-like capabilities, including everything from computer vision and natural language processing to deep learning and agentic AI.

At this stage, there may be as many ways to get AI wrong as get it right. But exploring enterprise AI in action can illustrate its potential, spanning a wide range of tasks, including:

• Speeding drug development
• Reinventing telecom business processes
• Designing toy cars
• Pollinating crops
• Increasing efficiency in large-scale manufacturing

Looking at what different companies are doing with AI—and how they’re doing it—can provide inspiration for others as they envision AI applications of their own.

Example 1: Goldman Sachs taps AI to write code

GenAI tools not only produce written language and images but also churn out computer code. Goldman Sachs is conducting a “proof of concept” for assisted coding tools powered by GenAI.

Even as industry peers J.P. Morgan, Citigroup, Deutsche Bank, and Bank of America were reportedly blocking their employees from using GenAI, Goldman Sachs was putting the technology in the hands of its programmers to determine its early value. CIO Marco Argenti explains that while it wouldn't be wise to immediately trust all important processes to AI, it's imperative that businesses try to envision its potential. The firm has several proofs of concept underway in areas that include software development and document classification, which have shown some promising initial results.

Taking things a step further, Goldman began testing an autonomous software engineer from AI startup Cognition in 2025 that is expected to soon work alongside the firm’s 12,000 human developers.

Takeaways

Goldman Sachs envisions AI as a companion to software developers rather than a replacement for them. It is also testing out large language models to upgrade the document classification management processes currently performed by traditional AI. The company is keeping the names of the tools it’s using under wraps, along with the specific departments that are testing the technology, but it has shared some early results.

• Developers have been able to use GenAI to write as much as 40% of their code automatically, which Argenti says could improve the productivity of human developers by somewhere in the low double digits.
• The bank is using the software not only to generate code but to test it as well.
• The large language models are at least as good as humans at managing the company’s millions of documents.
• Goldman Sachs’s CIO sees GenAI as one of the biggest disruptions of his lifetime, on par with the Internet, apps, or the cloud.
• The greatest challenges, he says, are adapting internal controls to mitigate any risks to the company or its customers and finding talent with large language model expertise.
• The company’s use of GenAI has expanded, with its GS AI Assistant now available to all Goldman Sachs employees after successful testing with approximately 10,000 staff members.
• Argenti says he sees the autonomous software engineer, named Devin, as a “new employee” who can start performing tasks on behalf of developers.

Learn more

• CIO Marco Argenti shared some early GenAI results in this Wall Street Journal¹  Q&A.
• Argenti tells CNBC how he envisions having hundreds—and, ultimately, thousands—of autonomous software engineers² working on behalf of the company, depending upon use case.
• In a recent podcast episode³, two analysts from Goldman Sachs research discussed the power of automating more day-to-day enterprise processes, such as software development.
• Reuters covers Goldman Sachs’ firmwide launch⁴ of its AI assistant in 2025.
• The Wall Street giant is “aggressively scaling AI⁵ for increased productivity and efficiency,” according to PYMNTS.

Example 2: Australian telecom company Telstra develops from AI assistants to agentic reinvention

In mid-2023, Telstra developed a pair of GenAI tools—Ask Telstra and One Sentence Summary—to boost customer support after service activation, part of a strategic five-year agreement with Microsoft Azure OpenAI Service. These tools help agents quickly access accounts, product details, and summaries of past interactions, enabling faster troubleshooting and more personalized care. The GenAI systems reduced follow-up contacts by 20%, with 90% of its customer service employees reporting time savings and improved post-sale support quality, according to Telstra.

“We’re unlocking value and growth as technologies like generative AI proliferate at speed,” says Kim Krogh Andersen, group executive, product & technology at Telstra, in an article on the Microsoft Web site, adding that the company was “on track to improve all key business processes with AI by 2025.” Indeed, the company took a bigger leap forward with the announcement of its joint venture with Accenture to use AI to reinvent business processes and develop agentic workflows.

Takeaways

As Forrester VP and senior research director Frederic Giron explains it, Telstra’s newer approach to AI points to a fundamental shift from a more piecemeal approach to automating existing tasks and business processes to automating isolated tasks along existing business processes to “reimagining how work gets done.” It addresses what Forrester has called the “process chasm”—the gap between a focus on productivity optimization and true business reinvention.

• Achieving a competitive advantage with AI requires more than rolling out GenAI assistants—it requires a full rethinking of how tasks are accomplished in the first place.
• Telstra is combining its crown jewels—its deep telecom expertise—with Accenture’s AI assets and talent to create AI solutions for everything from process optimization to customer experience.
• The approach is to develop AI solutions for internal use, perfect the system, and then productize it for sale.
• While other companies are juggling multiple partnerships, Telstra has winnowed its AI collaborators down to two, with an emphasis on co-innovation.

Learn more

• Forrester’s Giron explains how Telstra’s “build, then sell⁶” approach amortizes large substantial up-front AI development costs across a broader revenue base and also creates new revenue streams.
• Telstra executives explain the company’s development of specialized co-pilots⁷ for customer service agents.
• The Guardian reports that Telstra’s CEO, Vicki Brady, told investors that AI will play a big part⁸ in Telstra reinventing itself over the next five years and that its workforce would likely shrink by 2030.
• A press release on the Accenture website explains how the seven-year joint venture Telstra will use from Accenture’s $3 billion AI investment⁹ to reinvent business processes end-to-end through new capabilities like agentic AI¹⁰.

Example 3: Moderna’s data and AI application speed mRNA development

Biotech company Moderna’s AI investments have paid off for drug development at a time when speed was vital for marketplace success. Founded more than a decade before the COVID-19 crisis, the company spent years building an integrated data science and AI software to support repeatable development of thousands of different mRNA-based medicines and vaccines. The web-based application includes reusable code for workflow automation, data capture, and model-building. This helps scientists design novel mRNA constructs, improve their efficacy and order samples through a high-throughput, preclinical production line.

Digital and analytics-based from the get-go, Moderna was built to beat the odds of traditional high-risk, high-return pharma development. Drug companies typically spend a billion dollars developing a drug in the hopes of getting many billions in return, but see a success rate of less than 15%. AI can improve those chances to 50/50 and reduce the time-to-market, as well. The AI bet paid off for the Cambridge, Massachusetts, company, which was able to develop a top COVID-19 vaccine in record time, showing around 95% efficacy for prevention of illness from the virus, according to the U.S. Centers for Disease Control and Prevention. COVID-19 vaccine in record time, showing around 95% efficacy for prevention of illness from the virus, according to the U.S. Centers for Disease Control and Prevention.

Takeaways

While the company’s COVID-19 vaccine garnered the most attention, it is only one of Moderna’s products built using its AI and data science tools for mRNA development. The company approached it as a software capability. “For over a decade, Moderna has built a massive library of data, which is our biggest asset as a digital-first biotech company,” Brice Challame, the company’s VP, Data & AI Transformation, wrote in a 2023 blog post, “This library has allowed us to create our own integrated data ecosystem to fuel and iterate upon our algorithms.”

• Many companies get hung up on the initial infrastructure investment for an AI project. However, Moderna recognized the value that an application could deliver across multiple products in the long term.
• Early investments in cloud infrastructure, IoT, analytics, and automated processes created the foundation for AI work. “We relied on digitization early on, not for the sake of digitization but for generating data,” explained Moderna’s former chief digital and operational excellence officer Marcello Damiani in an article by the Digital, Data, and Design Institute at Harvard. “This allows us to build better algorithms, which helps build the next generation of medication,” Damiani says.
• An mRNA vaccine is itself an information-based product, a synthetic molecule that sends instructions to the body’s cells for training its immune system. So, Moderna built a web-based drug-design application to streamline the process of changing the information encoded in a synthetic molecule. Behind the scenes, numerous AI algorithms help inform decision-making, particularly in preclinical products, for which they provide data-centric predictions on the best sequences to code for a certain protein much faster than humans can. The software captures data throughout the process to produce new and better algorithms over time.
• The company also uses AI to improve clinical trial planning, quality control, and even its call center operations.
• Beginning in 2023, Moderna partnered with OpenAI to build its own instance of ChatGPT. Today, GPTs are embedded across Moderna's business functions—from legal, to research, to manufacturing, to commercial—as purpose-built assistants.

Learn more

• The Digital, Data, and Design Institute at Harvard offers background¹¹ on Moderna’s early implementing of AI.
• Moderna’s Brice Challame shares some lessons learned on the company’s “journey to becoming a real-time AI organization.”¹²
• MIT Sloan Management Review’s Me, Myself, and AI podcast interviewed¹³ Moderna’s former chief data and AI officer Dave Johnson about the role of AI in vaccine development.
• ZDNet describes¹⁴ Moderna’s drug-development technology as an “AI factory.”

Example 4: Mattel early-adopts AI image generator DALL-E for product development

GenAI screeched into the general consciousness with wide rollouts of AI language models capable of generating human-like text. Months earlier, toymaker Mattel began working with a generative image-creation tool, OpenAI’s DALL-E system, for creating realistic images and art based on natural language inputs.

Mattel has no plans to replace its product designers with artificial counterparts anytime soon. Instead, the toy maker is deploying AI to help designers come up with ideas for new Hot Wheels cars. More recently, Mattel inked a 2025 deal with OpenAI to develop AI-powered products and experiences based on Mattel’s brands.

Takeaways

Mattel’s GenAI exploration illustrates the potential benefits of human-machine collaboration in a creative field.

• Mattel began using the image generator in October 2022.
• Designers can type in a natural language request for a “scale model of a hot rod convertible,” and the system may deliver variations on a custom image based on a 1938 Ford in yellow with whitewall tires.
• The user can work with the image generator to explore different options—a color-change, a hardtop, a tweak in body type—to deliver a final rendering of a new toy car.
• GenAI allows designers to think more broadly, offering models or options they might not have conjured up alone. “Ultimately, quality is the most important thing,” says Carrie Buse, the director of product design at the Mattel Future Lab. “But sometimes quantity can help you find the quality.”
• The toy maker, which ran into some trouble a decade ago with its “Hello, Barbie” doll which recorded and uploaded children’s conversations to the cloud, said in a press release it will “emphasize safety, privacy, and security in the products and experiences that come to market” as a result of its newest OpenAI collaboration.

Learn more

• The Associated Press highlights¹⁵ Mattel’s early entry into GenAI.
• Microsoft Source details¹⁶ how human designers work with DALL-E to generate and refine new Hot Wheels designs.
• In the Harvard Business Review¹⁷, Thomas H. Davenport, who is Professor of IT and Management at Babson College, and Nitin Mittal, who is a principal at Deloitte Consulting, explore how GenAI will change creative work.
• PYMENTS¹⁸ digs into the 2025 Mattel-OpenAI collaboration on AI-enabled toys.

Example 5: Costa Group deploys computer vision-powered pollinators instead of bumblebees

If there’s any buzzing about in the tomato greenhouses of Australia’s Costa Group in Guyra, New South Wales, it’s not coming from bumblebees. Using the natural pollinators (such as bees) for indoor farming is illegal there—native honeybees struggle in covered environments. And for biosecurity reasons, Australia has long banned the import of nonnative European bumblebees, which are often used for greenhouse pollination in the Northern Hemisphere. Instead, the produce grower is using robotic pollinators—powered by computer vision—on one million tomato plants.

In summer 2021, Costa Group started using Polly robots from Israeli firm Arugga AI Farming. Early results showed the machines produced a 15% higher yield compared to manual pollination and up to a 7% higher yield compared to bumblebees.

Takeaways

For the Costa Group, the robotic pollinator has benefits beyond efficiency. While they’re not allowed to use bees (which is how most greenhouse pollination is performed), the agribot option is not only more efficient than manual pollination, but it also reduces the spread of viruses as it requires no human contact with the tomato plants.

• Tomato plants need to be shaken to self-pollinate, a task typically done by industrial bumblebees, or by workers rustling trellises or using vibrating wands.
• The Polly robot travels between the rows of plants, using computer vision and deep learning to identify which flowers are ready for pollination. On the back end, the Arugga system uses a streaming analytics toolkit to process raw video from greenhouses. The imagery is then annotated, and the model trains on the resulting datasets, which are updated monthly. The model improves over time and is exported to the robots, which can run the algorithm individually and are essentially edge computing devices.
• The robot then emits compressed air pulses to vibrate the flowers that are due, similar to the way bumblebees perform buzz pollination.
• Arugga’s AI is getting smarter. In 2024, it said it could pollinate one hectare (about 2.5 acres) with 3.5 robots, with expectations to reduce that to 2.5 robots thanks to an improved algorithm.

Learn more

• Costa Group’s AI-powered pollinators are just one example of the agricultural computer vision applications¹⁹ in an Imaging & Machine Vision Europe article.
• Produce Plus²⁰ explains how the robots determine which flowers are ready to pollinate.
• Arugga’s co-founder and vice president of business development explains how the company’s AI pollination, a replacement for manufacturer bees, can help growers increase their yields²¹.

Example 6: Procter & Gamble deploys AI, machine learning, and edge computing to streamline production

P&G is looking to digitalize and analyze data from its over 100 manufacturing sites. It’s developing AI capabilities, such as machine learning and computer vision, to maximize equipment health and availability, assess product quality in real time on the production line, and improve energy and water usage. The consumer-packaged goods maker has focused its early attempts on its paper products and baby care segments with pilots in the United States, India, Japan, and Egypt. An early project used AI to predict finished paper towel sheet lengths, thereby delivering the right amount of product to customers—just one of many efficiencies the company hopes to achieve.

“We want everything done in real time, on the line, with technology, so that we know instantly if something is wrong,” Chair, President, and CEO Jon Moeller explained in a discussion with Goldman Sachs Asset Management. “And with AI, we can start trending toward knowing instantly—knowing before it happens, not just when it happens.”

The company is also using AI in supply chain management. Previously, it required bringing together data from disparate software used for manufacturing, supply chain, marketing, quality assurance, and laboratory information systems, a tedious, manual task requiring hundreds of labor hours each time. Now it’s partnered with analytics provider phData to develop an AI-powered system using KNIME’s open-source analytics to automate data integration. Then it can analyze supply, demand, and inventory data in real time to forecast future inventory requirements and highlight potential supply chain disruptions.

Takeaways

P&G’s goal is smart manufacturing at scale, said former CIO Vittorio Critella, with a focus on predictive maintenance, predictive quality, and sustainability optimization. AI could help allow for fully touchless operations down the line.

• A digital enablement office is prioritizing product manufacturing, packaging use cases, and packaging processes that could be implemented organization-wide.
• The company can transmit data from production line sensors to the cloud to build and train machine learning algorithms that are then deployed back to the edge on the factory floor.
• P&G has piloted the use of AI to minimize over packing of paper towel rolls (by more accurately predicting finished sheet lengths) and using computer vision for real-time quality control for diapers and feminine pads (where the precise, high-speed assembly of multiple layers of materials is essential).
• Large volumes of complete datasets from its paper machines can be used to train machine learning algorithms to predict which machines require maintenance and pinpoint opportunities to lower energy usage.
• Machine learning and AI must be embedded into operations and culture to deliver value, P&G’s CIO says.

Learn more

• Consumer Goods Technology offers an overview²² of P&G’s digital system, which uses IoT sensors and AI.
• P&G’s former CIOs talked to Deloitte’s CIO Journal about the opportunities and challenges of deploying AI at scale²³ in manufacturing.
• CIO.com describes how P&G has been able to mature beyond AI experimentation²⁴ to scale increasingly sophisticated use cases.
• Microsoft Source posted news about its work with P&G²⁵ on digital manufacturing.
• CEO Moeller told Goldman Sachs²⁶ that he sees AI potential in “moving innovation from the lab bench to very sophisticated computers” to speed up molecular discovery, explore more areas for innovation, and come up with more successful product ideas.

Example 7: UK high-speed railway team builds a better construction plan with AI simulator

The High Speed 2 (HS2) railway running from London to Manchester is the biggest transport infrastructure project in the UK. Joint venture company Align formed a working group involving three international infrastructure companies and is responsible for the design and construction of one of the most challenging legs of the project: the so-called C1 section. This 21.6 km (13.42 mi.) stretch northwest of London includes the 3.4 km (2.11 mi.) iconic Colne Valley Viaduct and a ‘twin-bored’ tunnel of roughly 16 km (9.94 mi.). At the height of the five-and-a-half-year project, more than 1,500 people will be working on-site. Align invested in ALICE—an intelligent construction sequencing product—to model contingencies and devise optimized and realistic construction schedules for the HS2.

Preconstruction—the first phase of a project during which companies plan and schedule a job’s entire scope, estimate costs, and analyze needs—is a crucial stage. But much has changed over the past couple of decades in the critical path method that architectural, engineering, and construction companies use to plan their projects. It’s typically been a Herculean challenge to come up with one or two plans, given the effort required to build a schedule. Using AI, companies can churn out hundreds or thousands of options in a few hours, with a full analysis of their effect on cost and schedule.

Takeaways

Align already had a construction schedule in place that helped it win the C1 project but used ALICE to double-check its assumptions and look for opportunities to improve the plan for its viaduct substructure work. Scheduling a massive infrastructure project, with its complex interdependencies and constraints, is hard to do well. The construction simulator incorporates these interdependencies in an algorithmic equation that can analyze thousands of scenarios and review them based on the company’s goals.

• The construction simulator helps companies virtually build construction plans with a variety of different inputs before actual construction begins.
• Align used the construction simulator to create dozens of scheduling options in just 10 minutes.
• It then ran “what if” analyses on the software to decide how to build the viaduct more efficiently.
• AI can rapidly manipulate massive amounts of parameters that can affect construction—such as labor, equipment, and material availability, construction methods, and zoning issues—in ways that human planners alone could never do.
• Align replicated three years of planning work in just six weeks.
• More recently, HS2 inked a 2024 deal with Futuria to be its AI delivery partner in a newly formed consortium led by Atos, to implement AI-driven solutions for improving rail operations, streamlining construction workflows, and enhancing asset management.

Learn more

• Management consultancy Roland Berger describes how AI can increase efficiency throughout the construction value chain²⁷.
• Construction Dive covers a range of preconstruction AI applications²⁸.
• In 2022, the Align-ALICE partnership won the award for industry innovation²⁹ from the British Construction Industry Awards.
• Railway Technology explains how an Atos-led consortium will build digital twins³⁰ of HS2 infrastructure, with railway-specific knowledge from Arup and AI from Futuria.

Example 8: Thomson Reuters streamlines development of AI tools for journalists, lawyers, and compliance pros

Dating back to its 1851 founding as a news agency, Thomson Reuters has always been about data, in one form or another. Today, it delivers data-based business services in the realms of legal, tax, and accounting, and news. What is now its legal division introduced natural language search 30 years ago, and its R&D group has been testing and delivering AI-enabled innovations ever since.

In 2022, Thomson Reuters created an AI application to speed its machine learning innovation by implementing common data and model governance, and standardizing its model release process.

Takeaways

Thomson Reuters’s AI application provides not only a common workspace for AI oversight, but a system for managing AI-specific risk with the goal of balancing speed and governance. There are a host of challenges to effective AI model performance, such as the potential for algorithmic bias and changes in the distribution of data. These challenges only become more complex as companies grow their deployment of AI-enabled systems, eventually growing beyond the abilities of data scientists to manually track them over time.

• Thomson Reuters worked with Amazon Web Services to develop a tool that provided a standard “clickable” user interface that data scientists and model owners across the company could use without writing code.
• The application offers AI workspaces, data service, model registry, AI annotation, and monitoring for issues such as data or model drift and bias.
• The application streamlines the traditional performance modeling done by human data scientists, who monitor the model deterioration and work with the appropriate teams to make necessary adjustments.
• User adoption was the most important success factor and the biggest challenge. Agile development with weekly demos for users to gather feedback was key.
• The company’s portfolio of AI-enabled offerings helps human journalists spot trends, provides predictive mapping capability for tax compliance, and offers natural language queries and AI-powered search algorithms for legal, tax, and accounting pros.
• Thomson Reuters emphasizes human-in-the-loop AI to reduce errors and improve model performance.

Learn more

• Maria Apazoglou, vice president for AI, machine learning, and business intelligence, shared details on the AI platform³¹ with Deloitte’s CIO Journal.
• The AWS Machine Learning Blog offers a nitty-gritty look³² at the development of AI technology.
• In a 2025 article, Thomson Reuters explains how three levels of human-in-the-loop testing³³ by 2,500 subject matter experts helped create its GenAI assistant, CoCounsel, which can conduct in-depth research, analyze extensive and complex data, and generate diverse types of content quickly.

Read more