Thought Piece: AI agents that actually work

Hi there,

Half of US companies now use generative AI for marketing projects, according to Capgemini research. But here's the ROI problem: these tools can't handle complex workflows or large-scale data without constant babysitting. Try to justify that investment beyond pilot programs. Good luck.

AI agents represent the next phase of AI transformation. They're autonomous systems that handle tasks like supply chain management or financial analysis with minimal oversight. The result? You can actually measure the ROI through lower costs and better efficiency. No more explaining away budget overruns.

AI agents make decisions on their own and deliver business outcomes you can put on a spreadsheet. Companies in real estate, aerospace, energy, manufacturing, and finance are seeing returns that transform their innovation budgets from cost centers into profit drivers.

Gartner's 2025 strategic technology trends show that by 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024. Even more striking: 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028. Smart organizations are getting ahead of this curve to capture first-mover advantages.

AWS's latest research shows the momentum is building fast. Their 2025 study found that 60% of organizations globally have appointed dedicated AI executives like Chief AI Officers to accelerate adoption. Plus, 92% are planning to hire for roles requiring gen AI skills in 2025. The message is clear: executives see AI agents as essential for competitive advantage.

You'll need a team with data science, UX design, machine learning, and software development skills to build one properly. But here's the good news: the investment math has completely changed.

What makes AI agents different and more profitable

Think of AI agents as really smart assistants that make basic chatbots look primitive. They deliver returns that traditional automation can't match. An AI agent handles tasks with minimal human help while generating measurable business value.

Old-school automation breaks when something changes. AI agents learn and adapt, protecting your technology investments over time. The key difference is agency, and the ROI that comes with it.

Traditional AI responds to your prompts. Agentic AI takes initiative. It examines data, performs research, compiles tasks, and then takes action in the digital or physical world through APIs or robotic systems. This autonomy translates directly to cost savings and revenue opportunities.

Here's an example: you want to analyze market trends for a real estate portfolio. An AI solution takes your requirements, grabs data from your systems, and gives you a complete analysis for making decisions. When market conditions shift, it adjusts instead of crashing. The result? Faster decisions, better outcomes, and lower operational costs.

AI agents need very little human input once deployed. They learn as they go, understand context, and can read and change data from different sources. They get human language and make smart decisions based on what they've learned. All while reducing your labor costs and improving output quality.

Two main types:

Autonomous AI agents face customers and make high-level decisions. They handle customer questions and manage complex processes without human help. This directly improves customer satisfaction scores and reduces support costs.

Assistive AI agents work inside your company to boost what your team can do. Satya Nadella put it well: business logic is moving to agents that work with multiple databases, making systems much easier to swap out when needed. This protects your infrastructure investments.

How to build an AI agent that delivers ROI

Building AI agents needs structure, but development speed has changed dramatically. What took 100 days to build can often happen in one day with modern AI tools. More importantly, the cost structure has shifted in your favor.

1. Define your AI business strategy first 

   Write down exactly what you want your AI agent to accomplish, but start with the business outcomes. Document measurable goals: cost reduction percentages, time savings, error rate improvements, or revenue increases. This becomes your success framework.

2. Build your team economically 

   Get people who know data science, machine learning, UI design, and software development. Here's the good news: AI has made this more affordable. Effective digital solutions now cost six figures instead of seven. That's a dramatic shift in the investment equation.

   Organizations are taking a dual approach to AI talent. They're developing internal workforce capabilities while also recruiting externally. If you don't have all the skills in-house yet, consider a mix of training your current team and bringing in specialized expertise. This hybrid approach often delivers better ROI than pure outsourcing.

3. Pick your tech stack for flexibility and cost control 

   Choose programming languages, ML models, and APIs. Build everything so you can swap out pieces later as technology gets better. This protects your investment as the AI landscape evolves.

   Most organizations aren't going with one-size-fits-all solutions. While 40% plan to use AI models off-the-shelf, 58% are developing custom software using pre-existing models, and 55% will build applications on fine-tuned models using their own data. The key is keeping your data safe while achieving the performance you need and controlling costs.

4. Design for measurable outcomes 

   Map how data flows from start to finish, but include measurement points throughout. Decide if you want to build parts separately or all at once. Add ways for users to give feedback and, more importantly, ways to track performance metrics that matter to your business.

5. Get your training data ready 

   Use data from your operations, outside sources, and users. Label everything so the AI knows what each piece means. Clean up errors and inconsistencies. Quality data directly correlates to better AI performance and stronger ROI.

6. Build and train with monitoring 

   Set up your training environment. Split data for training and testing. Start your ML model and watch how it performs. Include business metrics alongside technical ones from day one.

7. Test everything against your ROI goals 

   Run unit tests, get real people to try it, and compare different versions. Check if it's accurate and fast enough, but also verify it's hitting your business targets before full deployment.

8. Launch and optimize for ongoing returns 

   Connect it to your existing systems. Monitor how it performs through user activity and feedback. Make improvements based on what you learn. This continuous optimization is where long-term ROI compounds.

Real-world ROI numbers that matter

AI agents are already transforming industries and delivering measurable returns that justify the investment.

Real Estate: AI agents predict market trends and manage client relationships through automated follow-ups. Major platforms analyze over 100,000 property listings to optimize investments. JLL saw deal closure rates jump 41% and cut negotiation time by 60%. Direct impacts on revenue and operational efficiency.

Aerospace: Boeing uses AI agents for predictive maintenance, analyzing equipment data to suggest when maintenance is needed. This prevents costly breakdowns and extends component life, turning maintenance from a cost center into a profit protector.

Manufacturing: Siemens deploys AI agents to optimize energy use, automatically adjusting operations for better efficiency. They've boosted performance by over 25% through energy optimization, quality control via computer vision, and automated compliance reports. Measurable gains across multiple cost categories.

Energy: Companies are implementing smart grid management systems that autonomously adjust power distribution based on demand patterns, reducing operational costs while improving grid stability.

Financial Services: JPMorgan Chase uses AI agents for regulatory compliance and risk management, analyzing transaction data to spot risks and ensure compliance requirements are met even during volatile markets. This reduces regulatory risk while improving operational efficiency.

The ROI timeline:

Smart organizations see returns in phases:

• Months 1-3: Pilot programs show initial efficiency gains and cost reductions

• Months 4-12: Scaled deployment delivers measurable operational improvements

• Year 2+: Compound benefits as agents learn and optimize, plus competitive advantages become apparent

What's next?

The companies that win won't just adopt AI fastest. They'll build systems that deliver measurable business value while working with people, not against them. The goal is human-AI collaboration where technology makes expertise stronger and ROI more predictable.

The shift is happening now. Gartner predicts that AI agent machine customers will replace 20% of interactions at human-readable digital storefronts by 2028. Organizations need to start planning for workflows designed around agentic AI, with humans added at high-value decision points.

Building AI agents is just one piece of a bigger digital transformation puzzle. Companies getting 100x returns know that real success means rethinking data systems and company culture too, but always with ROI as the north star. If you want to see how all these pieces work together, my new book, "100x", shows the complete framework organizations use to turn AI into a competitive edge.

The future of AI isn't just about having the smartest algorithms. It's about creating an innovation strategy that puts AI agents at the center of your competitive advantage while delivering measurable business outcomes.

Start small with pilot programs that focus on measurable impact. The time to plan your AI strategy is now. Waiting means watching competitors pull ahead while you explain budget overruns instead of celebrating returns.

The smartest move? Begin with one process that everyone hates doing manually and where you can easily measure improvement. Show results. Then scale from there. In the AI era, the ability to adapt means survival, but the ability to demonstrate ROI means thriving.

Much Love,
Matt

At Lighthouse, we love featuring fresh perspectives from our community of AI, tech, and innovation leaders. Got insights to share? Just reply to this email—I’d love to hear from you!