.png)
Imagine a world where your software doesn't just sit idle, waiting for your input, but actively works for you, like an invisible assistant, making decisions and executing tasks autonomously. This is not just AI hype. Enter: Agents, the next big evolution in software.
Agents are action-oriented programs that perform complex tasks without direct user intervention every step of the way.
- Dynamic Interfaces: Agents respond to multimodal inputs whether it's text, voice, or even a messy document.
- Automated Workflows: Agents save you time by interacting directly with your systems, databases, and other software on your behalf.
- Reinforcement Learning: Agents continuously improve through human feedback.
The clearest, most recent example of this fast-approaching future is Anthropic’s recent release of Claude Computer Use. Highly recommend you click through to watch the demo video to really appreciate what it's capable of and what it looks like from a 'future of UI' perspective. Obviously, this trend is only just getting started....
This is an example of an agentic product. The user submits a prompt (text, voice), and the Agent “does the work.” In this case, the Agent navigates and interacts with the App interface so the user doesn’t have to. How will product design change as a result of this kind of workflow automation?
For innovative companies, agents are the new apps. But it’s not just a new form factor. It’s a paradigm shift in how to approach product strategy, product design, and product development.
Agents vs Apps
Agents are different from traditional Apps in three fundamental ways:
1. Human-In-The-Loop UI
Apps are designed to help people create, read, update, and delete (CRUD) various pieces of data in various formats, over and over and over again. Agents are designed to automate repetitive, structured work like this, cutting down many manual steps in a Human Workflow, and therefore increase throughout and/or decrease headcount requirements.
However, there is still the need for Humans to not only “approve” AI-curated or AI-generated work in production but also improve it continuously by providing the Agent with clear and accurate quantitative and qualitative feedback. Therefore, the Agent UI is going to surface AI recommended actions for users to evaluate, edit, reject, or approve. This feedback will be used to improve future recommendations, so Human productivity gains grow in a compounding fashion over time through the regular use of Agents. Users will still be able to CRUD like they’re used to at all times, but the UI will be more collaborative as they share the space with the Agent, and need a one or multiple ways to interact with the Agent via multimodal LLMs. (See: Replit Agent for a good present-day reference.)
Agents can respond to natural language, text, audio, visuals, and code in large quantities and without over-burdensome data cleanliness in ways that a traditional App plainly cannot do. Users can express what they want however they want: 1) an audio note coupled with a website URL reference, 2) a video message and Google Drive folder, 3) a long-form document and a set of third-party APIs, or 4) all the above. The Agent processes the inputs; reasons for any Recommended Actions; surfaces these for Feedback & Approval; if Approved, programmatically uses any necessary UIs, APIs, codebases, and databases autonomously in order to execute the Actions; and then synthesizes the Outcomes into a best-fit final response.
For this reason, Agents are more dynamic, expressive, and forgiving than traditional Apps.
Business Impact: This shift demands that product and tech leaders embrace new thinking. This product design space has hardly been explored yet, and of course some artifacts from today's paradigm will evolve and live on in the next, but whatever comes next will surely look, respond, and interact unlike any traditional App we've seen before.
2. The Agent Journey
Agents work by configuring an LLM model to identify certain questions or requests as Recommended Actions, and then triggering the relevant Action protocol if Approved. Depending on the use case, Agents may need access to proprietary data or authorization to interact with relevant APIs.
Agents take Actions by interacting with internal and external APIs, writing their own API calls, retrieving API values, exchanging specific data and potentially changing state across one or multiple systems.
This is mapping the Agent Journey, and it's the new Service Blueprint. This is what innovators are focused on now, and they need data-centric product leaders, data engineers, and AI engineers to design, build and scale these solutions.
Business Impact: Agents require data-centric product leaders proficient in APIs and collaboration with data engineers and AI engineers. Businesses need modern data infrastructure and automation-minded leaders to capitalize on this emerging opportunity.
3. Reinforcement Learning from Human Feedback
The method for driving continuous improvement with Agents is vastly different and more powerful than with traditional Apps. To optimize Agents over time, focus on refining prompts, ‘scoring’ responses, and driving Reinforcement Learning from Human Feedback -- both explicit (e.g. thumbs up/down, open-text feedback) and implicit (e.g. low engagement, 'failed' sessions) -- back into the model's training automatically for compounding improvements, validated by data-driven benchmarking.
Business Impact: B2B software success metrics will shift to how effectively Agents deliver 'outcomes', most likely with as few human interventions as possible.
Agents: The Next Era of B2B Software
Agents and the broader Service-as-Software trend are set to become the next evolution of B2B software and professional services. High-growth companies today already focus on data-centric solutions and operational efficiency. Agents embody this drive and take it to the next level.
We are building Agents in-house and for our clients today with our expertise in AI, data, and custom software development. You can create a competitive, compounding advantage in the market for your business when you start investing in Agents today.
Your expected short-term and long-term gains include:
- Modernized Data: Streamline and automate your data pipelines, and make your data more accessible through visualizations and natural language (i.e. a ‘digital brain’ for your company.)
- Supercharged Productivity: Automate routine work to boost productivity and reduce inefficiencies in the workforce.
- Accelerated Innovation: Create more business value more quickly and affordably across practically any organizational dimension such as R&D, Go-To-Market, Growth, and Operations.
Get in touch with Rootstrap about an AI Innovation Audit for your business, and other ways we can accelerate your roadmap and drive innovation today.
