Success requires more than bolting AI onto old processes; it demands rethinking solutions to age old problems, integrating AI into existing systems, reimagining workflows, building strong data foundations, and aligning governance across teams.

Exciting product developments from global technology companies are heralding a new age of agentic AI for enterprises. One of the latest advancements has made a big splash on the market: OpenAI launched GPT-5, designing a unified, multimodal AI model with advanced reasoning, long context, and persistent memory.

This follows the release of Claude Skills API and Google A2A framework, along with powerful agents like ChatGPT Agent, and ‘Deep research’, which can autonomously execute complex tasks: browsing, summarising, building slides, and completing orders. For enterprises and AI users, this marks the beginning of agentic AI at scale, less prompting, more doing, bringing smarter automation into workflows like customer operations, legal, and R&D.

Industries are responding in kind to AI innovation. In Box’s The State of AI in the Enterprise report, 60% of companies expect to achieve AI transformation within two years. And while 87% of companies have started using AI agents of any kind, 41% are using agents for advanced, fully autonomous operations that are delivering much higher productivity gains.

How can companies make the most of these innovations? This is a critical question to keep top-of-mind, especially as at least 30% of GenAI projects will be scrapped post-pilot, derailed by bad data, weak risk controls, rising costs, or fuzzy business value by the end of 2025, Gartner warns. Success in the future of industry rides on avoiding these AI pitfalls.

Building a clear AI strategy

OpenAI’s new GPT-5 model and ChatGPT agent are significant developments for enterprise AI – steps towards automating complex tasks from start to finish. For leaders, it’s important that excitement around automation must be combined with an-all round clear AI strategy.

While we’re seeing bold progress and ambition in innovation, modern GenAI tools are still in their infancy. They are still prone to making mistakes and hallucinating, which pushes the correction workload back to the user. In fact, statistical modelling shows that even with a low per-step error rate, the compounding effect adds up: after just five tasks, there’s a 37% chance the agent fails – creating a significant risk of user churn.

What’s clear in the market is that buzz is stoking demand, which can outstrip the ability to implement the technology. This is where leaders encounter challenges. Unlocking real value from AI requires unified orchestration, not isolated deployments. Without it, AI’s impact stays constrained.

Organizations therefore need to establish and put proper guidelines in place before emerging technologies can be successfully onboarded. Pausing and figuring out an AI-first strategy is what will make “iterative, collaborative workflows” successful, where process engineering remains as important as ever.

Crucially, maintaining human oversight throughout AI-related changes, such as interoperability across AI tools and pre-existing work systems will be mission-critical. Across corporate structures, we’re seeing the lines increasingly blur, because AI workflows are bringing multiple competencies together.

Teams can now do the work of adjacent functions. Through capabilities from AI agent tools, we will see a redistribution of human resources to where they are most effective.

Becoming an AI-first company

With all the promise of AI’s great impact, it’s important to unpack the specific areas where we are seeing AI alleviate the burden of menial work. Integrations in industry are demonstrating that enterprise IT is shifting from app-centric stacks to multiagent architectures, where fleets of AI agents work together across systems to achieve shared goals.

In the supply chain, AI is helping to monitor inventory, spot shortages, and trigger supplier orders, all without bespoke integrations. McKinsey has identified that technology leaders will deploy these capabilities in three main ways: through super platforms with built-in agents ready to plug in, AI wrappers that connect internal systems with third-party services securely, and custom agents fine-tuned on proprietary data for tailored use cases.

Together, these models mark a fundamental reimagining of how enterprise technology is built and operated. Yet before AI is even considered as a solution, you need to interrogate your core challenges.

If you don’t have a clean process today, it’s very hard to generate impactful automation. When the process is ironed out, and you’ve needled out the ‘Why?’ for AI, consider long-term how multiple tools and departments work in concert.

Mapping out AI success

There is no ROI in rushing AI onboarding. Becoming an AI-first company means progressing through maturity stages, from AI supporting insights, to automating workflows, to agents collaborating, self-improving, and eventually running business functions end-to-end under human oversight.

Success requires more than bolting AI onto old processes; it demands reimagining workflows, building strong data foundations, and aligning governance across teams.

Companies must also invest in change management and foster a culture of experimentation and trust. Ultimately, AI-first is both a technology and mindset shift, where humans and machines collaborate to transform how work gets done.

Yashodha Bhavnani is Head of AI, Box