In 2025, you’ll find a curious contradiction in most boardrooms. Nearly every Fortune 500 company has launched a generative AI initiative, often with splashy headlines and hefty budgets. But behind closed doors, many of those same executives admit they haven't seen much impact. According to a recent McKinsey report, almost 80% of companies are using generative AI in some way, yet close to the same number say it hasn’t changed their bottom line.

That’s not because the technology is flawed. It’s because most companies are still thinking of AI as a feature or plugin, something that adds a little speed here, a little automation there. The organizations that are seeing real returns are taking a different approach. They’re not sprinkling AI on top of old processes. They’re rebuilding those processes from the ground up around autonomous agents.

These agents don’t just answer questions or generate text. They’re designed to act. They retrieve information, take actions, and make decisions within guardrails. They behave more like coworkers than tools. Think of them as AI teammates who can handle routine work, collaborate with others (including humans and other agents), and continuously learn from context.

And here’s the real shift: the companies embracing agents aren’t just in tech. Leaders across finance, retail, telecom, and manufacturing are redesigning how work gets done by placing AI agents at the center of operations. In this post, we’ll explore how Fortune 500s are going “AI native”, not just adopting AI, but rewiring their businesses for it.

Along the way, we’ll look at real examples, share lessons from early adopters, and offer a glimpse of what the future of enterprise workflows might look like when your AI doesn't sit on the sidelines but runs with the team.

Who’s Leading the Shift?

If you think AI agents are only being deployed at startups and Silicon Valley labs, think again. Some of the world’s largest and most established companies are already using them to reshape operations, boost productivity, and reduce costs. What’s interesting is that they’re not just plugging in chatbots. They’re building, or buying fully capable digital teammates to support employees, interact with customers, and make decisions in real time.

Here are some of the most notable players leading the transition to AI-native operations:

Microsoft: AI That Works Where You Work

Microsoft isn’t just building AI agents for others—it’s using them internally across its own workflows. The company has embedded its Copilot agents inside Microsoft 365, GitHub, Dynamics, and other platforms. These agents don’t require users to leave familiar tools like Word, Excel, or Teams. Instead, they add AI into the flow of work, where it can quietly speed things up without disruption.

For example, GitHub Copilot helps developers write code faster, with productivity boosts up to 55 percent in some teams. In Microsoft 365, Copilot summarizes emails, drafts documents, and helps teams prepare presentations. The company’s internal sales teams use AI agents to summarize customer data and prioritize outreach, projecting a 6 percent uptick in close rates just from using smarter prep tools.

And Microsoft isn’t keeping this in-house. Today, over 85% of the Fortune 500 use its AI products. That means Copilot-style agents are showing up everywhere: from HR teams automating onboarding to legal teams scanning contracts. The common thread? AI agents aren’t a new tab to open. They’re a coworker in the tools people already use.

JPMorgan Chase: Turning Advisors into Super-Advisors

In finance, JPMorgan is treating AI agents as a force multiplier. The bank’s generative AI assistant, Coach AI, supports more than 25,000 wealth management professionals by retrieving research, summarizing news, and even generating tailored investment suggestions.

The results are eye-catching. Advisors using the tool now find information up to 95 percent faster. During market swings, AI prepares updates in real time so that advisors can be more proactive. The bank credits these changes with helping deliver a 20 percent year-over-year increase in sales in its asset management division.

Beyond wealth management, JPMorgan is integrating AI agents into hundreds of operational workflows. From reviewing legal contracts to detecting fraud, these agents are doing work that previously took entire teams. The firm expects its number of AI use cases to double this year from 450 to over 1,000 powered by a $17 billion technology investment.

Walmart: Building a Team of Specialized Agents

Walmart may be known for physical stores, but it’s rapidly becoming one of the most sophisticated AI-native enterprises in retail. The company recently introduced a generative AI shopping assistant that acts more like a concierge than a chatbot. Behind the scenes, it’s actually a team of coordinated agents: one handles product recommendations, another checks availability, another manages checkout and delivery.

Each of these agents is powered by a retail-specific large language model trained on Walmart’s proprietary data. That allows them to understand inventory, store layouts, and customer preferences at a granular level.

Internally, Walmart is using AI agents to speed up product development, spot fashion trends, and optimize supply chains. One tool, called Trend-to-Product, helps designers go from idea to shelf 18 weeks faster than traditional methods. Meanwhile, in customer service, AI agents resolve common issues while escalating complex ones to human reps with a detailed context summary.

The company is also preparing for a future where shoppers have their own AI agents. Walmart expects a world where your personal assistant negotiates directly with theirs, and they’re already optimizing for “agent-to-agent” commerce.

AT&T: Autonomous Assistants Behind the Scenes

Telecom is another sector seeing quiet but dramatic transformation through AI agents. AT&T has deployed autonomous digital assistants in fraud prevention, customer support, software development, and network operations.

One such agent monitors transactions for suspicious activity and stops fraud in real time without needing human approval. In customer service, AT&T orchestrates several agents to handle billing questions, upgrades, and technical support in parallel—then surfaces a smart set of options to human reps. This approach reduces call times and improves upselling.

Even on the technical side, developers are using multi-agent systems to write, test, and validate code in a supervised environment. Think of it as a digital scrum team: one agent writes the feature, another checks it, and a human engineer oversees the whole thing.

What’s consistent across AT&T’s deployments is structure. The company ensures every agent operates with oversight and clear accountability. Agents don’t run wild. They run with purpose.

Others: Agents Across Every Industry

It’s not just tech, finance, and retail. Companies across sectors are embracing AI agents in specific, high-value areas:

• Bank of America’s Erica has handled over 2.5 billion interactions, helping customers transfer funds, check balances, and get answers quickly.

• Accenture has committed $3 billion to building out AI-powered services and agent ecosystems for clients across industries.

• Professional services firms like PwC are rolling out ChatGPT-style assistants to over 100,000 employees to help with research, analysis, and tax work.

These examples highlight a growing trend that how companies are not experimenting anymore instead incorporating in their core workflows.

What AI-Native Enterprises Do Differently

It’s one thing to plug AI into an existing tool. It’s something entirely different to rebuild a process so that AI sits at the center. The most successful enterprises aren’t just adding agents; they’re redesigning how work flows, how decisions are made, and how humans and machines collaborate.

After looking at how Microsoft, JPMorgan, Walmart, AT&T, and others are deploying AI agents, a few clear patterns start to emerge. These patterns tell us what it really takes to become an AI-native enterprise.

1. Build Specialized Agents That Work Together

Rather than creating one giant AI to do everything, most companies are building smaller, specialized agents each with a clear job. For example, Walmart’s shopping assistant isn’t just one model answering all questions. It’s a set of agents: one recommends products, another checks store availability, another manages payment. Each agent is trained for a specific role, just like a human team member.

Then comes orchestration. These micro-agents pass information back and forth behind the scenes to support a single, fluid experience. AT&T takes a similar approach. When a customer calls in, one agent might handle billing while another checks your upgrade eligibility. A lead orchestrator stitches everything together and delivers a ready-made solution to the human rep.

This modular setup allows companies to upgrade or fine-tune agents without overhauling the entire system. It’s scalable, flexible, and reduces the risk of any one model going off the rails.

2. Train Agents on Your Own Data and Context

Public models like GPT-4 are incredibly powerful, but they’re generalists. In practice, the best results come when agents are fine-tuned on industry- or company-specific data.

Walmart’s agents are trained on retail-specific information like product catalogs, pricing logic, and customer shopping behavior. JPMorgan’s Coach AI has access to proprietary market research and investment data. These agents aren’t just smart, they’re familiar with the business.

That’s what makes them trustworthy. By grounding agents in real enterprise data, companies ensure they provide accurate, context-aware answers. More importantly, they can act on that knowledge confidently, whether it’s flagging a fraud alert, recommending a product, or adjusting a supply chain decision.

3. Embed AI in the Flow of Work

One of the smartest things Microsoft did with Copilot was integrate it directly into the tools people already use. There’s no extra login, no clunky interface. You write an email in Outlook, and Copilot suggests edits in real time. You open Excel, and it offers to summarize your sales trends.

This model, embedding AI directly into core systems, is becoming the norm. AT&T’s agents are plugged into its operations stack. Walmart’s agents interact with inventory databases and product design tools. The idea is to make AI less of a sidekick and more of a built-in collaborator.

By placing agents in the flow of work, companies get higher adoption, better data access, and more consistent results. And for employees, it means AI is just... there. Quietly helping. Making everything a little faster, a little easier.

4. Make Agents Proactive, Not Just Reactive

Most people are used to AI as a tool you prompt. You ask, it answers. But AI agents don’t have to wait for instructions. When designed well, they can monitor systems, detect problems, and take action before humans even notice.

In supply chain operations, for instance, some agents now automatically reorder stock when inventory runs low. In IT, agents can detect unusual traffic patterns and shut down potential threats before they escalate. In finance, agents surface unusual spending patterns or credit risks without being asked.

This shift from reactive to proactive is one of the biggest advantages of agent-based systems. It turns business processes into living systems that adjust in real time, much faster than any human team ever could.

5. Keep Humans in the Loop, By Design

Autonomy doesn’t mean removing humans entirely. The best AI-native companies know where to let agents run and where to put human guardrails in place.

At AT&T, for example, even though AI handles large chunks of code writing or customer issue resolution, a human is always in the loop to review and approve actions. At JPMorgan, advisors use Coach AI to generate ideas and recommendations, but they’re the ones delivering advice and making final calls.

This structure builds trust. Employees feel empowered, not replaced. And the organization stays in control of outcomes. It’s also a practical safety measure. When agents operate in critical or regulated environments, oversight isn’t optional, it’s essential.

6. Redesign the Whole Workflow, Not Just a Step

Perhaps the most powerful thing AI-native enterprises do is take a fresh look at entire workflows. Instead of asking, “Where can we insert AI?” they ask, “How would this process work if we designed it from scratch with AI in mind?”

This often leads to simpler, faster systems. Instead of three teams passing information across handoffs, an agent can handle the end-to-end flow. A customer places an order, the agent checks inventory, processes payment, and schedules delivery, all without human input unless something goes wrong.

Companies that rethink processes at this level often see dramatic results. One Fortune 500 manufacturer redesigned its supply chain workflows around AI agents and reported a 32% reduction in inventory costs and $12 million in annual savings within the first year.

The takeaway? AI-native companies aren’t just adopting tools. They’re architecting a new way of working. And it’s not a small tweak, it’s a full-system redesign, built from the bottom up with agents in mind.

Build or Buy? How Enterprises Are Getting Their Agents

Once a company decides to go AI native, the next question is: how do you actually make it happen? Do you build everything in-house? Buy from vendors? Or stitch together both?

In practice, most Fortune 500 companies are choosing a hybrid approach. They combine external AI platforms with internal data and customization. The result is a balance between speed, control, and scalability.

Let’s break down how the best in the business are building their agent stacks.

Buying the Foundation

Many companies start by using off-the-shelf AI tools. These include platforms like Microsoft Copilot, OpenAI’s APIs via Azure, or enterprise products from vendors like Moveworks, UiPath, or Aisera. These platforms come with prebuilt agents for common workflows: customer service, HR, IT support, and more.

Using these tools gives companies a fast path to value. There’s no need to hire a huge AI team or build models from scratch. Instead, business teams can quickly deploy and test agents using no-code or low-code interfaces.

For example, Microsoft’s Copilot can be rolled out across Office 365 with minimal IT lift. Teams instantly gain AI capabilities inside tools like Word, Outlook, and Teams, drafting documents, summarizing threads, even generating slides from meeting notes.

And this isn’t limited to general use cases. There are industry-specific solutions too. Healthcare companies use AI agents to handle patient triage. Banks use them to flag fraud. Retailers use them to resolve product inquiries. Each of these starts with a vendor solution that’s then customized for the enterprise.

Beam AI: A Platform for Agent-Native Operations

One example of a vendor purpose-built for this shift is Beam AI. The platform provides plug-and-play AI agents that can automate everything from customer support to scheduling, data entry, and procurement workflows.

What sets Beam apart is its focus on agent orchestration. Instead of offering a single assistant, Beam lets companies deploy multiple coordinated agents, each with a specific task, working together in complex enterprise workflows.

Need an invoice processed? One agent pulls the data from a PDF, another checks it against a purchase order, a third submits it to the ERP system. It’s like having an invisible back office team that works 24/7, without burnout or coffee breaks.

Beam is designed to integrate with your systems, not replace them. Whether it’s Salesforce, ServiceNow, or SAP, Beam agents plug in and act just like a human would, only faster, and at scale.

For companies that want the benefits of autonomous operations without rebuilding their entire tech stack, platforms like Beam provide a practical and reliable on-ramp.

Building In-House for Strategic Control

While vendor platforms get you started quickly, some enterprises choose to build their own AI capabilities, especially when they want more control over the data, decisions, or competitive edge.

Take Walmart, for example. The company developed its own retail-specific large language model, trained on years of proprietary shopping, inventory, and pricing data. That model now powers its internal and customer-facing agents, giving Walmart full ownership over how recommendations, fulfillment, and service decisions are made.

JPMorgan has also hinted at building its own models for financial use cases, even filing trademarks for a product called “IndexGPT.” Bloomberg went a step further and launched BloombergGPT, a 50-billion-parameter model fine-tuned for finance.

The main reason to build? Differentiation. These companies want agents that behave exactly the way their business demands. And they don’t want to share that intelligence with anyone else, not even a cloud provider.

Of course, this comes with tradeoffs. Building custom AI requires top talent, serious infrastructure, and a long runway. Most companies won’t take this path for every use case. Instead, they might build their own agents for high-value functions (like trading or product design), and use vendor solutions for everything else.

Don’t Forget the Plumbing: LLMOps and Governance

Regardless of whether they build or buy, every enterprise faces the challenge of managing their agents at scale. That means handling costs, performance, updates, and risks.

This new field often called LLMOps (large language model operations) is all about creating the infrastructure to support agents safely and reliably. It includes:

• Prompt versioning: so you can track how an agent’s instructions evolve

• Access controls: to ensure agents don’t touch the wrong data

• Monitoring: to catch when something breaks or goes off-script

• Audit trails: to document what decisions the agent made and why

Enterprises are also deciding where to host their AI. Some prefer the flexibility of cloud providers like Azure or AWS. Others, especially in regulated industries, choose on-premise deployments to keep everything in-house.

A growing number of companies are building internal “agent hubs”, centralized platforms that teams across the org can use to deploy and monitor AI agents, share prompts, and standardize best practices.

Think of it as the IT helpdesk, but for AI coworkers.

In short, becoming AI native doesn’t require reinventing everything. But it does require making smart decisions about how you source, manage, and scale your agent workforce. The companies doing this well treat AI like any other major workforce investment: with a mix of speed, strategy, and structure.

How AI Agents Are Changing the Enterprise From the Inside Out

The move toward AI-native operations is not just a tech upgrade, it’s a full rewire of how organizations function. When you start giving software the power to take action, the ripple effects touch everything from team structure to budgeting to performance management.

Here’s how autonomous agents are starting to reshape the enterprise from the inside.

New Org Charts, Fewer Hand-Offs

Most business processes were designed around human limitations. You needed different teams for each step sales passed things to finance, finance to ops, ops to legal, because no one person could handle everything.

AI agents break that pattern. Since agents can operate across systems, respond in real time, and handle multiple tasks in sequence, there’s less need for siloed hand-offs. One agent can handle a workflow from start to finish, escalating only when it hits an edge case.

This allows companies to flatten their orgs. Some managers now oversee both people and agents. Others are giving employees more autonomy by letting them “hire” their own agents to support routine tasks.

For example, a product manager might launch an agent to prepare reports, respond to support tickets, and monitor key metrics. Instead of needing a full team, they just need the right assistants.

From Hiring Humans to Hiring Agents

Here’s a question more teams are asking: should we hire a person for this job, or should we spin up an agent?

This isn’t about replacing people across the board. It’s about choosing the right tool for the task. Repetitive, data-heavy, and rule-based work is ideal for agents. Creative, interpersonal, and strategic work still belongs to humans.

But the math is changing. Companies are starting to factor agents into their workforce plans. Rather than hiring another 50 people to process claims or review documents, a company might deploy an agent army that costs a fraction of that, and runs 24/7.

A Fortune 500 manufacturer that used agents to manage inventory and logistics saw a 32% drop in carrying costs and saved $12 million in the first year. That didn’t just reduce headcount pressure, it freed up people to focus on supplier negotiations and higher-value planning.

Productivity Without the Burnout

Every organization wants to do more with less. AI agents are starting to make that possible in ways that don’t require longer hours or larger teams.

For example, British Columbia Investment Management Corporation found that employees using Microsoft Copilot saved 30% of the time on internal audits and reported a 68% increase in job satisfaction. In other words, AI made work not just faster, but less frustrating.

In development teams, GitHub Copilot has helped engineers write code 55% faster on average, allowing teams to ship more features without increasing headcount. And in sales, some organizations are seeing close rates improve with the same number of reps—thanks to agents who prep calls, suggest follow-ups, and write outreach emails automatically.

The big insight here is that productivity is no longer about working harder. It’s about letting agents handle the repetitive parts so humans can focus on the creative and interpersonal work that really moves the needle.

Redefining Roles, Creating New Ones

The rise of AI agents doesn’t just reduce work. It changes it.

Many employees are finding that the most boring parts of their jobs, chasing approvals, copying data, writing status updates, are now automated. That gives them time to think, plan, and problem-solve.

At the same time, entirely new roles are emerging. Companies now need people who can manage, design, and refine agents. These include:

• Prompt engineers who shape how agents think and respond

• AI product owners who manage agent behavior just like a software product

• Agent QA analysts who check outputs and handle edge cases

• Workflow designers who combine humans and agents into seamless processes

Some companies are also training “AI coaches” to help colleagues get the most from new tools. These aren’t engineers, they’re usually power users from the business side who act as internal champions.

It’s a shift in mindset. Employees aren’t competing with agents. They’re working alongside them.

Rethinking Cost Structures

One of the less talked-about benefits of agents is what they do to your budget. When agents can handle a significant chunk of the work, companies don’t need to scale their teams in a linear way.

That’s especially powerful in high-volume environments like customer service, finance, and logistics. AI doesn’t just save time, it changes how much it costs to serve a customer, process a transaction, or ship an order.

It’s not all savings, though. Enterprises are also seeing new costs: cloud compute fees, AI vendor licenses, and investments in oversight and security. But in many cases, the tradeoff is worth it.

For example, Microsoft’s own research found that companies deploying generative AI through Copilot saw an average of $3.70 in return for every $1 spent. That’s not a science experiment—that’s bottom-line value.

Measuring What Matters

With agents in the mix, traditional metrics need an update. Companies are moving beyond headcount and throughput to track things like:

• Tasks completed per agent

• Human oversight rates

• Average time saved per workflow

• AI utilization by department

• Error reduction from automation

These new KPIs reflect a blended workforce, one where people and agents share the load. They also help companies pinpoint which agents are delivering value and where fine-tuning is needed.

The best part? These metrics often surface wins that aren’t obvious. Like the legal team that resolved contracts 3x faster after deploying an AI summarizer. Or the sales team that re-engaged dormant leads thanks to an agent that spotted patterns in CRM data.

Once you start measuring agent impact, you start finding unexpected leverage all over the business.

Keeping It Trustworthy: Ethics, Oversight, and the Human Factor

Scaling AI agents across a business isn’t just a technical challenge—it’s a trust challenge. As agents take on more responsibility, organizations need to be sure these systems are operating safely, fairly, and transparently.

The companies doing this well don’t treat AI governance as a bolt-on. They build it into the design from day one. Here's how.

Fairness and Bias Are Everyone’s Problem

AI agents learn from data. If that data is biased, so are the outputs. And when agents are making decisions, recommending who gets a loan, how a customer complaint is handled, or what product is shown first, that bias can have real consequences.

Responsible enterprises are putting guardrails in place. Before rolling out an agent, they run fairness audits. They check whether results vary unfairly by race, gender, location, or income. They adjust or retrain the model if the answer is yes.

Companies like Microsoft and Google have published responsible AI guidelines. These aren’t PR statements. They’re internal playbooks for how to build AI systems that reflect company values and minimize harm.

Human in the Loop: A Non-Negotiable

Even the best agents will sometimes make mistakes. That’s why smart enterprises build in human checkpoints, especially in high-stakes areas like healthcare, finance, or legal.

JPMorgan, for instance, has humans reviewing all sensitive outputs from Coach AI before they reach clients. AT&T gives engineers final say before agents make network changes. At Walmart, complex customer support issues are still escalated to human reps with AI offering a full context summary to speed things up.

This model works because it matches strengths. AI agents handle the repetitive grunt work. Humans deal with nuance, empathy, and exceptions.

Over time, the goal isn't to remove humans from the loop entirely. It’s to put them in the right part of the loop—supervising, refining, and improving the agent team just like they would a human one.

Documentation and Audit Trails Matter

In regulated industries, you can’t just let an agent act and hope for the best. You need records. You need to explain what the system did and why.

That’s where auditability comes in.

Companies are now designing agents with built-in logging. Every decision, prompt, and output is stored. If something goes wrong, you can trace it. If a regulator asks how a loan was denied or a purchase was flagged, the company has receipts.

This kind of traceability also helps teams improve agents over time. You can analyze failure patterns, optimize prompts, and roll out updates just like you would with a customer service rep or backend system.

It’s not about micromanaging AI. It’s about maintaining control and accountability.

Protecting Data at Every Step

AI agents often need access to sensitive data. That’s what makes them useful. But it also creates risk.

Leaders are addressing this through a mix of policy and infrastructure. They’re choosing vendors that offer end-to-end encryption, on-premise deployments, and role-based access control. They’re minimizing how much data agents see, and ensuring agents can’t wander into areas they shouldn’t.

For example, a finance agent might access invoice records but be blocked from payroll data. A customer service agent might know your order history but not your health info.

Some companies are even exploring synthetic data training agents on generated examples rather than real customer files to reduce exposure.

Data privacy isn’t just a compliance issue. It’s a reputational one. And the companies that take it seriously earn the long-term trust of both customers and employees.

Building Trust Inside the Organization

The biggest threat to AI adoption? Fear. Employees worry about being replaced, micromanaged, or forced to use tools they don’t understand.

That’s why leading enterprises are investing in AI literacy, training, and change management. They’re creating safe spaces for people to experiment with agents, ask questions, and give feedback.

Many have launched internal AI academies. Others are appointing AI champions within each department to support their peers. Some run regular “agent office hours” where employees can get help setting up or optimizing their tools.

The result is a cultural shift. AI stops being something done to people, and starts being something done with them.

Over time, this approach doesn’t just build trust. It creates a workforce that’s more empowered, more curious, and more open to change.

The Road to an AI-Native Enterprise

By now, it should be clear: going AI native is not about slapping an assistant on top of your existing tools. It’s about rethinking how your company works from the inside out—powered by agents that learn, adapt, and act with purpose.

This transformation is already happening at some of the world’s most recognizable companies. Microsoft is embedding copilots into every layer of knowledge work. JPMorgan is equipping advisors with real-time AI researchers. Walmart is building shopping agents that collaborate like a team. AT&T is using autonomous assistants to detect fraud, support customers, and even write code.

And behind the scenes, companies are redesigning workflows, investing in new platforms like Beam AI, and setting up governance structures to ensure their agent workforce is safe, fair, and effective.

So what can your enterprise learn from all this?

1. Start With Use Cases That Matter

Don’t launch AI pilots just to say you’ve done it. Pick workflows where agents can save time, reduce errors, or drive revenue. Look for processes that are structured, repetitive, and high-volume. That’s where agents thrive.

2. Build the Right Mix of Tools and Talent

Use vendor platforms to get quick wins, but also start building internal skills. You’ll need people who understand prompts, workflows, oversight, and metrics. Think of it as creating an internal agent ops team.

3. Rewire, Don’t Just Repaint

Use this moment to rethink outdated processes. Ask: If we had an AI teammate from day one, how would we have designed this workflow? The answer is often simpler, smarter, and faster than what you're doing now.

4. Put Humans in the Right Place

AI isn’t here to replace people. It’s here to level them up. Let agents handle the busywork so your team can focus on what humans do best—judgment, creativity, empathy, and leadership.

5. Make Trust a Core Feature

Governance, security, ethics, and transparency aren’t nice-to-haves. They’re essential. Build them into your agent deployments from day one. It will pay off in user adoption, regulatory compliance, and brand reputation.

Final Thought: You’re Not Too Late

Becoming AI native doesn’t require a moonshot. It starts with a mindset shift.

Think of AI agents the way you thought about cloud computing in the 2010s or mobile apps before that. They’re not a side project. They’re the new foundation. And the earlier you start building on that foundation, the more leverage you create over time.

Most enterprises today are still dabbling. The leaders? They’re rewiring. They’re changing how work flows, how decisions are made, and how value is created.

That’s the AI-native edge.

And now’s the time to build it.