The AI Dilemma: Structure Kills Innovation, Agility Creates Chaos (Part 2 of 3)

Two Contrasting Approaches – Are Both Wrong?

In the first article of this series, we examined the Structured Approach to AI Implementation — a method that emphasizes meticulous planning, data governance, and risk management. While it offers order and consistency, we explored how its rigidity can stifle innovation, slow progress, and leave organizations unable to adapt to emerging opportunities. Despite its strengths, the Structured Approach often proves to be the "wrong" way to build effective AI solutions in today's dynamic environments. (Check out Part 1 here)

Now, we turn our attention to the Agile Approach, which takes the opposite stance. This approach prioritizes speed, adaptability, and rapid experimentation, offering a way to innovate quickly and respond to changing conditions. On the surface, it seems like the perfect antidote to the drawbacks of a Structured Approach. However, the Agile method brings its own set of challenges. By focusing too heavily on flexibility and real-time learning, organizations risk falling into chaos, wasting resources on unfocused efforts, and undermining long-term sustainability.

This second installment in the series dives into the Agile Approach—what it gets right, what it gets wrong, and why it, too, might not be the optimal way to implement AI solutions across an organization. Once again, the goal is not to dismiss the value of these methodologies but to understand their limitations and the potential pitfalls they introduce when taken to the extreme.

In the final article, we’ll explore how finding the right balance between structure and agility can lead to a more effective, sustainable approach to AI implementation.

The Agile Approach to AI implementation

The Agile approach emphasizes speed, adaptability, and a willingness to embrace uncertainty. Unlike traditional, rigid methodologies, this approach encourages rapid experimentation, continuous iteration, and the freedom to pivot based on real-world feedback. It’s about deploying AI models and solutions quickly to learn from their performance in practical settings, using those insights to drive continuous improvement. This method fosters an environment of innovation, where teams are empowered to think creatively, experiment with new ideas and technologies, and remain flexible in the face of evolving business needs and technological advancements.

Central to the Agile Approach is the idea of decentralization, where small, cross-functional teams operate independently and make their own decisions about tools, platforms, and project priorities. The goal is to remove barriers and bureaucracy, enabling faster decision-making and more tailored solutions. Additionally, rather than relying on formal training programs, this approach promotes community-driven learning, where employees gain skills through engagement with AI communities, open-source projects, and external expertise.

However, while the Agile Approach can drive rapid progress and provide a competitive edge, it also comes with its own set of challenges. The emphasis on speed and flexibility can lead to a lack of strategic alignment, disorganized efforts, and even potential ethical oversights if governance is not prioritized early on. The reliance on informal learning may result in uneven expertise across teams, making collaboration and scaling more difficult. Furthermore, the absence of structured processes can sometimes create chaos, with duplicated efforts and inefficiencies. As we dive deeper into each component of this approach, we’ll see how prioritizing adaptability over structure can often be the wrong choice, creating significant obstacles that organizations must be prepared to address to ensure sustainable and effective AI deployment.

As in the Structured Approach we have identified 10 components for a fully Agile Approach to AI Development:

1. Start with Experimentation and Embrace Uncertainty
2. Prioritize Opportunistic Data Usage
3. Focus on Agile Development Over Rigid Planning
4. Empower Small, Cross-Functional Teams
5. Invest in Community Driven Learning
6. Deploy Quickly and Adjust in Real-Time
7. Minimize Governance and Compliance Until Necessary
8. Learn from Mistakes (“Learnings”) Publicly and Transparently
9. Focus on Culture Over Process
10. Scale Based on Grassroots Success

Let’s take a look at this each of these, let’s figure out why an Agile approach may just be the ‘wrong’ approach.

1. Start with Experimentation and Embrace Uncertainty

In rapidly evolving environments, embracing a culture of experimentation can drive innovation and uncover unexpected opportunities. The Agile Approach encourages teams to adopt a "fail fast" mindset, allowing for rapid testing of AI concepts without waiting for perfect conditions or fully developed strategies. By launching pilots with imperfect data and continuously iterating based on real-world feedback, organizations can quickly learn what works and adapt their methods accordingly. This approach fosters flexibility and accelerates the development of impactful AI solutions, making it possible to pivot and refine strategies on the fly.

Adopt a "Fail Fast" Mindset

Instead of setting concrete objectives and identifying use cases beforehand, companies should jump right into experimentation. Test out different AI concepts quickly to see what resonates and delivers unexpected value.

Pilot Without Perfection

Don't wait for a perfect data set or a fully thought-out strategy. Run pilots even with imperfect data to uncover opportunities and learn through doing.

WHY THIS IS WRONG

Diving into experimentation without concrete objectives can accelerate progress, but it also risks creating chaos and wasting resources on unfocused efforts. While embracing uncertainty allows for quick learning, a lack of strategic direction may result in half-baked projects that fail to deliver real value. Sometimes, a bit of upfront planning can prevent costly missteps and ensure efforts are aligned with business goals.

2. Prioritize Opportunistic Data Usage

In AI development, waiting for perfect data can significantly delay progress. The Agile Approach emphasizes using whatever data is readily available to kickstart AI experimentation, allowing teams to learn and adapt quickly. By letting real-world outcomes and experimentation drive the data strategy, organizations can identify gaps and prioritize data improvements based on practical needs. This method fosters a culture of data-driven discovery, where insights emerge from hands-on experience with AI models rather than extensive preliminary assessments, enabling quicker innovation and responsiveness to new opportunities.

Use What You Have

Instead of spending time assessing data quality and availability upfront, use whatever data is immediately accessible. Let AI experimentation drive the data strategy, revealing what’s missing or needed.

Data-Driven Discovery

Experiment with AI models to learn more about your data itself, rather than treating data assessment as a separate preliminary task.

WHY THIS IS WRONG

Using whatever data is readily available without assessing its quality can lead to unreliable AI models and flawed insights, undermining the credibility and effectiveness of AI initiatives. Relying on opportunistic data can also make it difficult to identify and address data gaps proactively, resulting in models that underperform or generate misleading results. This approach may create more work later, as teams scramble to fix data-related issues that could have been avoided with an initial quality assessment.

3. Focus on Agile Development Over Rigid Planning

To stay adaptable and responsive, the Agile Approach prioritizes rapid iterations and flexibility over rigid, comprehensive planning. By starting with small, experimental AI projects, teams can quickly learn from real-world outcomes and continuously refine their strategies. This method emphasizes making adjustments on the fly, allowing the AI development process to evolve based on practical insights rather than pre-set milestones. By focusing on speed and adaptability, organizations can stay ahead of the curve, react to changing circumstances, and deliver impactful AI solutions faster.

Iterate Rapidly

Rather than building a comprehensive AI roadmap, start with small, experimental projects that can be rapidly iterated upon. The focus should be on speed, adaptability, and flexibility.

Make Adjustments on the Fly

Use real-world outcomes to continuously adjust your AI strategy instead of sticking to pre-set milestones and KPIs. Let the results of early experiments shape your next steps.

WHY THIS IS WRONG

Prioritizing rapid iterations and making adjustments on the fly may seem efficient, but it can lead to projects lacking direction and clear objectives. Without a comprehensive roadmap, teams may struggle to align on priorities, resulting in fragmented efforts and wasted resources. This approach can also make it difficult to measure progress and success, as constantly shifting strategies undermine long-term planning and accountability.

4. Empower Small, Cross-Functional Teams

The Agile Approach emphasizes decentralization and empowers small, cross-functional teams to experiment with AI independently. By giving teams the freedom to select their own tools and platforms, organizations can foster innovation and speed up the development process. This approach encourages spontaneous and organic collaboration between tech and business units, allowing teams to tailor AI solutions to their unique needs. The result is a dynamic and adaptable environment where teams can quickly iterate and respond to emerging opportunities, driving rapid advancements in AI capabilities.

Decentralized Efforts

Allow different teams across the organization to experiment with AI independently, including selecting their own tools and platforms without waiting for centralized approval or alignment.

Organic Collaboration

Promote spontaneous and organic collaboration between tech and business teams, giving them the freedom to decide which technologies work best for their specific projects.

WHY THIS IS WRONG

Allowing small teams to independently choose their own tools and platforms can accelerate innovation and adaptability, but it also risks creating a fragmented ecosystem with incompatible technologies. This lack of standardization can make it challenging to scale successful projects across the organization and may lead to duplicated efforts and inefficiencies

5. Invest in Community Driven Learning

The Agile Approach to developing AI expertise emphasizes flexibility and adaptability through community-driven learning. Instead of relying solely on formal training programs, teams are encouraged to engage with AI communities, attend meetups, and explore open-source projects. This fosters a culture of continuous learning and allows for the rapid acquisition of new skills as the field evolves. Additionally, leveraging external experts on a short-term basis provides targeted insights without committing to long-term hiring, enabling the organization to stay agile and responsive to emerging AI trends and challenges.

Community-Driven Learning

Instead of formal upskilling programs, encourage your team to join AI communities, attend meetups, and learn from open-source projects. The goal is to create a culture of ongoing learning and exploration.

Leverage External Expertise

Bring in experts as needed on a short-term basis rather than making long-term hiring decisions too early in the process.

WHY THIS IS WRONG

Relying heavily on informal, community-driven learning can result in a lack of standardization and uneven expertise among team members. This inconsistency can hinder collaboration and lead to gaps in critical skills needed for successful AI implementation. Without structured training, teams may also miss out on foundational knowledge, relying instead on fragmented and potentially unreliable sources. While efficient in the short term, this approach can make it difficult to scale AI capabilities effectively across the organization.

6. Deploy Quickly and Adjust in Real-Time

The Agile Approach prioritizes speed and adaptability in AI deployment, encouraging teams to launch AI models as soon as possible to gain valuable real-world feedback. Rather than waiting for extensive testing, this method emphasizes getting models into production early and treating them as evolving systems that can be continuously refined. By embracing real-time experimentation, teams can quickly learn from user behavior and external changes, making necessary adjustments on the fly. This approach allows organizations to stay responsive and rapidly iterate, keeping AI initiatives aligned with current needs and opportunities.

Deploy AI Solutions Early

Don’t hold back AI models for extensive testing. Get them into production quickly and gather real-world feedback. This real-time feedback loop can be more insightful than controlled tests.

Embrace Continuous Experimentation

Treat AI models as living, evolving systems. Adjust them in response to user behavior and external changes without a formalized monitoring process.

WHY THIS IS WRONG

Deploying AI solutions quickly without extensive testing might seem like a way to gain rapid insights, but it exposes the organization to significant risks, such as unreliable model behavior, potential financial losses, or reputational damage. Relying on real-time feedback without a solid monitoring framework can also make it difficult to identify and resolve issues efficiently. While speed is tempting, a lack of initial testing and structured monitoring can lead to costly consequences that could have been avoided with a more cautious approach.

7. Minimize Governance and Compliance Until Necessary

In an effort to accelerate AI development, the Agile Approach advocates for starting with minimal governance and compliance measures. Rather than building comprehensive frameworks from the outset, this method suggests addressing ethical and regulatory concerns reactively, based on real-world needs as they arise. By adopting a light-touch approach, organizations can move forward quickly and iterate on their governance and ethical policies as practical challenges present themselves. This flexible strategy allows teams to innovate without being overly constrained, while still adapting governance practices to ensure responsible AI use as solutions scale.

Light-Touch Governance

Start with minimal governance and compliance efforts, addressing ethical and regulatory concerns reactively rather than proactively. As AI solutions scale, governance structures can be built around real-world needs rather than hypothetical scenarios.

Iterate Ethics

Test AI ethics policies in practice rather than investing significant resources into them upfront. Adjust ethical guidelines based on practical challenges that arise during implementation.

WHY THIS IS WRONG

Starting with minimal governance may speed up AI initiatives, but it’s a risky gamble that can lead to ethical missteps and costly regulatory consequences. On the other hand, implementing heavy, complex compliance measures from the outset can stifle innovation and delay progress. A pragmatic approach is to establish lightweight yet effective governance that can scale as needed, ensuring responsible AI use without slowing everything to a crawl.

8. Learn from Mistakes (“Learnings”) Publicly and Transparently

The Agile Approach embraces a culture of openness and continuous learning by encouraging teams to share their mistakes and learnings transparently across the organization. By fostering an environment where failures are seen as opportunities for growth, companies can accelerate collective knowledge and drive innovation. This method emphasizes gathering insights from setbacks and using those lessons to inform and improve future AI initiatives, allowing the organization to remain adaptable and relevant. Open experimentation and transparent communication help create a culture of trust, collaboration, and continuous improvement.

Open Experimentation Culture

Encourage sharing failures and learnings openly across the company to accelerate collective knowledge.

Continuous Insight Gathering

Use lessons learned from failures (and successes!) to create new AI opportunities rather than sticking to a rigid plan that may no longer be relevant.

WHY THIS IS WRONG

Sharing mistakes openly can foster a learning culture and accelerate progress, but it also makes failures highly visible, potentially damaging trust and providing competitors with an advantage. On the other hand, keeping mistakes under wraps may feel safer but slows down collective improvement. A balanced approach involves being transparent internally to drive learning while managing external communications carefully to protect the organization’s reputation.

9. Focus on Culture Over Process

The Agile Approach emphasizes building a strong cultural foundation where AI adoption grows organically and innovation thrives. By prioritizing cultural buy-in, organizations can empower employees at all levels to see the potential of AI and contribute their ideas. Crowdsourcing AI project concepts from both technical and non-technical teams can bring fresh and unconventional opportunities to the forefront. This method fosters an environment of collaboration and creativity, where the enthusiasm and insights of the workforce drive AI initiatives, keeping the organization adaptive and forward-thinking.

Cultural Buy-In First

Instead of investing in a top-down, structured AI initiative, foster an innovative culture where AI adoption grows organically from employees who see its potential.

Crowdsource Ideas

Gather AI project ideas from all levels of the organization, even from non-technical teams, to surface unconventional opportunities.

WHY THIS IS WRONG

Prioritizing culture over process can lead to inconsistent execution and a lack of accountability. Without structured guidelines and a clear framework, AI initiatives may become disorganized, resulting in duplicated efforts or projects that fail to align with broader business goals. This lack of cohesion can make it difficult to scale successful projects and ensure reliable outcomes across the organization.

10. Scale Based on Grassroots Success

The Agile Approach to scaling AI solutions emphasizes organic growth driven by proven value and enthusiastic adoption. Rather than imposing top-down mandates, successful AI projects are allowed to spread naturally through word-of-mouth and genuine user support. This method encourages flexibility, with strategies that can be continuously adapted as the AI landscape evolves, avoiding rigid, long-term commitments. By scaling only when projects have demonstrated their effectiveness, organizations remain nimble and responsive, ensuring that resources are allocated to initiatives with real impact and staying open to new opportunities for innovation.

Scale Organically

Only scale AI solutions that prove themselves valuable in practice. Let successful projects spread through word-of-mouth and enthusiastic adoption rather than by mandate.

Adapt Continuously

Be willing to pivot strategies as the AI landscape evolves, with no long-term commitment to a particular direction.

WHY THIS IS WRONG

Gradually scaling AI solutions may seem safe, but it can be too slow to capture emerging opportunities and may hinder an organization's ability to stay competitive. By waiting to expand successful projects, companies risk losing momentum and delaying the full benefits of AI adoption. This cautious approach can also frustrate teams eager to implement AI-driven improvements across more areas of the business, slowing down overall innovation and impact.

The Agile Approach to AI implementation

This approach offers a compelling alternative to traditional, structured methods by prioritizing flexibility, speed, and a fail-fast mindset. It provides organizations with the ability to rapidly adapt to changes, experiment with new ideas, and iterate based on real-world feedback. This approach can be incredibly effective in fast-paced environments where the ability to seize emerging opportunities can be a significant competitive advantage. By empowering small, cross-functional teams, embracing community-driven learning, and deploying AI models quickly, companies can keep their innovation cycles moving and continuously refine their strategies.

However, as attractive as this approach may seem, it’s not without its flaws. The very elements that make Agile so powerful—rapid experimentation, minimal governance, and decentralized decision-making—can also lead to significant drawbacks. Without a clear strategic direction, AI projects risk becoming fragmented and uncoordinated, potentially wasting resources on initiatives that don’t deliver meaningful results. The lack of formalized processes and governance structures can expose organizations to ethical pitfalls and regulatory non-compliance, especially as AI models become more integrated into core business operations. Moreover, relying too heavily on informal learning can create skill gaps and inconsistencies that hinder collaboration and scalability.

In the end, while the Agile Approach offers a pathway to quick wins and adaptability, it can also be the wrong strategy if not carefully managed. Organizations must recognize that the lack of structure can lead to chaos, inefficiencies, and missed opportunities for sustained growth. Striking the right balance between agility and thoughtful planning is key to maximizing the benefits of this approach while mitigating its inherent risks. As we consider both the Structured and Agile methods, it becomes clear that each has its strengths and weaknesses, and the true challenge lies in finding a strategic blend that aligns with the organization’s goals and long-term vision.

In the next part we will discuss how a much more balanced approach to AI implementations across the organization is more than probably the way to go.

What have been your experiences with implementing AI Solutions?

About me: I am the SVP of Digital Strategy at OZ Digital Consulting. If you would like to discuss creating AI solutions for your company, please contact me - that's what my team and I do for our clients. Message me about how we can do this for you.

I recently hosted the OZ AI Summit, which you can see a recap of here: https://followoz.com/best-of-ai-future-summit-2024/