By David Wang and Joel Deer
When Chattgpat launched at the end of 2022, it was a weekend call for many companies. For us, it was not just a signal, it was a catalyst. It endorsed the long -standing ambitions of our engineers and product leaders so that we solved the problems of consumer worries in which we struggled to crack with previous generation technologies.
Traditional chat -based interfaces, while reacting tasks are useful, often struggle to stay associated with user goals, handle multi -stop reasoning or take meaningful action. They can be like colleagues who look forward to the next assignment that expects needs.
Now, a new chapter is emerging when the agent takes the AI center stage. We have access to tools that not only responds, but also act. These agents can translate complex goals, plan multi -stop tasks, adapt to real time, and process workflows along with human professionals.
In professional domains, the evolution of AI has demanded a platform that prefers confidence, accuracy and domain skills. The values we have spent years to integrate into our system. Our work includes agent capabilities in products such as the Co -Council tax, tax and accounting, and the legal industry's first professional grade agent AI research tool.
The council can translate complicated purposes, plan and implement multi -stopwork flows, and can provide precision results like an experienced lawyer, an accountant or compliance officer. It is based on the industry's leading tools-including Westla for legal research, practical law for practical guidance, and check-outs for tax and accounting skills, allowing it to do real work, such as using a calculator to complete tax returns.
This means that instead of providing information, professionals can assign full assignments – such as working through tax returns or drafting legal movements and examining it – knowing that work will be handled on industry standards. Thousands of legal, tax and compliance experts behind these abilities are deeply guided by the experts, ensuring that the results are not only technically accurate but also linked to the way experienced practitioners are working.
Given this change as Chief Product Officer and Chief Technology Officer at a company to provide reliable skills to professionals, as the Chief Product Officer and Chief Technology Officer in a company to provide reliable skills to professionals, it will never be clear and our products will never be clear.
This change is especially important in high steaks domains that we support, including law, tax, compliance and risk. In these industries, accuracy, transparency and trust are the most important. The AI should perform reliable performance, be kept align with the rules and support the human decision.
Transfer from results to answers
Three years ago, our AI journey looked very different. We initially used to adopt Gut Hub's co -pilots like tools, and today, more than 80 percent of our engineers use them weekly. But it was just the beginning. Now, we are focused on complete agent systems – tools that can refer to internal documents, interact with servers, recover direct data, fix the tragedy and insects, and even create applications from the beginning.
Unlike the software code, which has a test and verified results, many expert domains have acceptable answers, which are better than others. Human decision is important at this place.
We have learned that embedded the AI engineers within the domain expert teams accelerates repetition and confidence. Our more than 250 AI engineers work together with more than 4,500 domain experts, including lawyers, accountants and compliance leaders, to create AI in real -world capabilities.
Imagine a system for lawyers that not only suggests clauses in the agreement, but compares documents, indicates legal risks, and increases complex issues for expert decision. Or a device for tax professionals that go ahead with the tax code flagging risks, adopting real -time data, and completing multi -stop workflows. These are not abstract ideas. They are embedded, results -based systems that are starting to explain how professionals are working.
Intelligence is just a half war
Tech culture has long been celebrating fast and breaking things. But in the law and taxes, breaking things is not an option. Pace is important, but confidence is even more valuable. It does not matter how much the skills of modern agents get, if professionals cannot trust them, they will not be adopted. The intelligent system needs to go beyond the results to provide transparency, consistency and monitoring.
Effective agent product designing is just as human challenge as technical products. The agent system needs to know when to increase the decisions, how to explain their reasoning, and how to adopt without misguiding user standards.
Human loop controls are central to this process. Experts guide development, stress testing edge issues, and ensure performance in the context that are the most important.
When the system is given the option of planning and implementing, small errors in the goals, contexts or data standards can lead to important errors. When it is clearly the most important importance, controlling automation without clear guards can remove human decisions.
Some powerful features of the agent system are hidden, such as their ability to maintain context, many sources, or decide when not working without enough information. These are the ones who allow professionals to work with more confidence.
Rebuilding system and teams
In our leadership characters, we focus on adaptation. We value curiosity, instead of just skill, the ability to learn quickly, and the cross -disciplinary work.
We have reorganized for small, highly affiliated teams and empowered articles experts to create AI behavior. We quickly repeat without sacrifice of hardship or confidence.
The future of real agent capabilities is not about the fastest or extremely autonomous system. This is about making the most useful: a system that can help professionals reliably in a moment where stake is high and time is short. The most meaningful agent AI will extend to the professionals who can get, especially when the error margins are low.
David Wang is the Chief Product Officer of Thomson Writers, where he guides product management, editorial, content and design, and product analtics teams. He oversees Thomson Reuters' global software and business information services portfolio product strategies and product development, including creating new AI software and technology for company professional users. Wang has over 15 years experience with business to business software, information services, and machine learning and AI system construction. Earlier, he served as a senior product leader in Facebook. He has also taken over senior roles in Nelson and has worked as a management consultant at Mac Cancery and Company. Wang holds a degree in engineering science from Toronto University and is an inventor with four nominated patents.
Joel Heron is the Chief Technology Officer at Thomson Writers, where he guides product engineering and AIR & D in legal, tax, audit, trade, compliance and risk. He joined themeson Reuters in 2022 through the acquisition of the Theate Trace, where he served as a CTO. Since then, it has helped to change Thomson Writers' technology strategy. By leading TR Labs and AI, their teams launched seven generative AI products in just 18 months, including AI for legal research, tax research and draft contract. Deer holds a master's degree in mechanical engineering from Texas University in Austin and has a bachelor in engineering from Texas Christian University.
Example: Dom Gazman
Be the latest with the current funding round, acquisition and more with the daily cranch base.
