Shifts and Frictions in Enterprises in the Era of GenAI Transformations
It has been many months since I shifted back to a field engineering role, where I have the opportunity to partner with enterprise customers to build new agentic systems that transform their business operations.
Along the way, I’ve made observations that completely changed how I had imagined the transformation playing out — some of them contrary to mainstream public opinion. Here I’m not leaning on the technical design side, but rather sharing what I’ve seen on the ground around change management and mindset shifts.
70% Efficiency, 30% Innovation — the economic diffusion problem
Most of the “ideas” I see for leveraging GenAI are about making the current manual approach faster and more automated: automating data loading, ML model building, dashboard generation, and so on. These don’t fundamentally change how value is created toward the end goal.
I wish there were more honest recognition from organizations that this efficiency work is part of the transformation. Often I see a culture in some enterprises that denies this effort is even a priority — there isn’t even an acknowledgment that “in order to stay competitive, we must think about how we can be more efficient. Only when we free up resources can we reallocate them to areas where we can innovate.”
Unless there is innovation — building systems that create higher value with GenAI — the resources freed up will not diffuse into economic value. Based on my observations, only about 30% of the effort is genuinely on innovation, which I define as creating a new system capable of generating a better output toward the goal. In a B2C business, this means value to consumers that they are willing to pay for.
Innovation share of GenAI effort
Drag to see where effort goes — and whether freed capacity creates new value.
Outcome: At roughly 30% innovation, reallocated time can start to diffuse into economic value — matching what I see in the field.
The gap in technical expertise and process to innovate
Even where innovation initiatives exist, most are high-level, loosely crafted concepts of how GenAI could change the business. These goals lack designs grounded in technical expertise, and as a result, most initiatives sound like a moonshot mission.
Contrary to some opinions, I don’t believe technical roles should be reduced on the assumption that the “business team” can do it all with GenAI. At least for the current phase, there actually needs to be more investment in technical teams and expertise building, so that business leaders can successfully translate their goals into outcomes.
Aside from designing a technically sound architecture, a big portion of the success formula is helping teams understand that this is an experimental process of rapid testing — and being candid about its limitations.
Take the example of an enterprise trying to build a chatbot that helps its sales team get faster insights and recommendations. Common gaps include expectations like:
- “We can automate this end to end without any human in the loop.”
- “The AI agent should be able to tell me why my forecast is off.”
Ambition
“We can automate this end to end without any human in the loop.”
Realistic today
Human review on high-stakes outputs; the agent proposes, humans approve or correct.
Good harness
Clear tool boundaries, evals, and an upgrade path as models improve — without a full redesign every release.
These are not the wrong ambitions. But the internal tech team often lacks the knowledge of how this is actually done — what a good harness can realistically achieve with the current level of intelligence, while still being able to scale without frequent major redesigns as the models get upgraded.
One mental model that helps: think about how to build many small innovation factories within the company that roll up to the broader business goals.
Anxiety about job security and role shifts is real
I have seen top-down pressure to significantly automate current technical workflows — think about the data science model-building process. This requires the individuals who hold that expertise and knowledge to share what they feel is their intellectual property with the organization. Complex ML model building, personal decision matrices, and the like are now being asked to be written down for GenAI automation.
In some cases, middle and senior managers have kept access to GenAI coding tools to themselves or a smaller group. This further creates distrust among junior members and adds resistance to seeing the transformation as a positive outcome for them.
Agile companies have an unprecedented advantage, and large enterprises are on a timer
I’m fortunate to work in an agile organization with high energy and a strong innovation culture, where people are empowered with the tools to innovate. I see great momentum in how fast the company is building new features and changing the way we deliver value to customers — in the go-to-market organization as well. Better demos, better business understanding, and better recommendations.
I see many startups now able to reset and completely rethink the best approach and org structure for this new era. Large enterprises, by contrast, can’t simply change, because the organization may not be able to keep operating through it. I believe many are on a clock: smaller, agile companies will be able to accelerate so much faster now, chipping away at the advantages and markets of the larger, more rigid players.
Many say there is a software-as-a-service apocalypse coming. I don’t fully agree — but I do believe many large companies, not just SaaS ones, will be replaced if they aren’t able to transform successfully.
These are observations from the field, and they’re evolving. I welcome recommendations and counterpoints — reach out if any of this resonates or if you see it differently.