Follow Datanami:
December 4, 2023

Pilot vs. Co-pilot: How Startups Will Reshape the Future of Work with AI

Jake Nibley


There’s been much debate about the future of work in an AI-driven world. Will the technology replace us all or fuel the next hyper-growth stage of the tech industry? The jury is still out, but it’s more likely that AI will fundamentally change how humans work, not eliminate jobs entirely.

Entrepreneurs and incumbents are now thinking about how to bootstrap AI into existing products and services, and how to augment human work with automation. The pervasive question in our industry is the following: how do we AI-ify everything? The answer is nuanced. AI can be used for different operations and take different forms; it can either be your co-pilot or your pilot.

The reality is that the future of work will involve AI and software will take a pilot-first approach. But this requires us to rethink how work gets done, not necessarily who or “what” does it.

Pilot vs. Co-pilot

The difference between co-pilot and pilot applications lies in the role of humans. Co-pilot applications improve user productivity by automating tasks in a project or workflow that a human oversees. Pilot applications are deeply embedded into workflows, systems of record, and tools needed to get work done and can complete these workflows end-to-end.

Jasper, which is an AI writer for marketing content, is a great example of a co-pilot application. Jasper uses generative AI to learn a company’s voice, tone, and style to help employees draft emails, product descriptions, and social copy that matches their company’s brand voice and ethos. In this case, workers are writing the copy to begin with and/or reviewing the copy created by the AI; the AI is augmenting the workflow, not replacing it.

Not every use case or workstream will lend itself well to a pilot-first approach, and it’s critical to have an evaluation framework in place to identify its use. The two most important factors to consider are tolerance to error and the level of process and human reasoning required.

Will AI be in the left seat or the right seat? (AI generated/Shutterstock)

Tolerance to Error

A high tolerance to error (i.e., typically associated with less risk) is more suitable for a pilot, while a low tolerance to error is more suitable for a co-pilot. High tolerance scenarios are areas where AI can begin to plug in immediately and play the role of pilot with limited human intervention and oversight. The pilot operates relatively autonomously, knowing that even if there are errors, a catastrophic outcome or failure is not likely. Meeting assistance tools, such as Microsoft Business Chat, are excellent examples of pilots. These tools can attend meetings, capture complete transcriptions, send summaries, track open questions and actions, and ensure that post-meeting follow-ups take place. These tools pilot many of the rote administrative tasks that surround meetings.

Alternatively, higher stake scenarios where errors must be avoided call for a co-pilot approach. This is especially true in industries like healthcare, where generative AI tools can be useful for streamlining a clinician’s workflow, like turning a patient interaction into clinical notes. Yet doctors can’t risk mistakes made by generative AI tools, so co-pilots will always be needed to keep a “human in the loop” to avoid everything from HIPPA violations to life-or-death misdiagnoses.

Level of Process and Reasoning Required

For tasks supported by a set of strict rules or guidelines where intuition or human reasoning isn’t needed, the AI pilot can be trained to follow a predetermined process with perfect precision. Examples of where an AI pilot shines are in accounting or legal applications. Because they’re both underpinned by specific rules like generally accepted accounting principles (GAAP) or case law and precedent, you can train the AI to follow these sets of rules with precision, in a deterministic way.

But some business tasks demand a human touch, be it intuition, emotion, or specific cultural or contextual understanding. These decision-making processes are perfect for an AI co-pilot. Here, the AI augments or enhances human creativity but doesn’t completely take over the process. This is where conversational AI agents like ChatGPT shine by supporting marketers’ creative ideation process, like generating examples of titles for a blog post based on drafted content or crafting ideas for social posts.

How We Get to a Pilot World

The fear of AI completely taking away our jobs is flawed. AI will certainly automate elements of our work, which will give us space to focus on aspects of our jobs that require human reasoning and other skills that computers cannot replicate. We are still very early in this AI journey, and the vast majority of the examples we see today are AI co-pilots. As we look ahead, our North Star should be software development with a pilot-first approach. To get there, we have to change how we think about the work being done.

(Berit Kessler/Shutterstock)

How do we realize this? Mega caps and SaaS incumbents are well positioned to develop co-pilot AI scenarios, with existing systems of records and systems of actions serving as a base for Al enrichment. Pilot scenarios take users out of incumbents’ existing products and are disruptive to their businesses. Startups with the right strategy can outmaneuver well-established incumbents in scenarios that merit pilots by focusing on customer needs and pain points to identify opportunities to automate and streamline existing processes using Al.

There have been many recent industry tailwinds that help co-pilots in the near term and pilots in the long term. The explosion of autonomous agent platforms, such as Fixie and Steamship, are helping the industry move toward enabling more pilot scenarios. Tools like Gorilla LLM enable deeper software integrations. By using natural language to describe the API endpoint that you need, it’s automatically created.

OpenAI’s investment in the plugin ecosystem easily equips models with tools (e.g., Wolfram Alpha, search, PDF readers, access to academic journals, calculators, etc.) so you can do more than just language generation and break into processes and complex workflows. Finally, mega caps are investing in platforms to enable model training and fine-tuning to produce domain or use case-specific co-pilots, opening the door for great pilot scenarios to emerge.

Everyone is going to become a manager of AI in this new way of working. This technology shift will be like going from walking to flying a fighter jet overnight. Startups in this space can take a fundamentally new approach to software and solving problems, getting back to first principles for how you drive business outcomes, irrespective of existing processes. Developers and entrepreneurs need to think about how they can break all preconceived notions of how work gets done and do that with AI. The time for disruption is now!

Jake Nibley

About the author: Jake Nibley is a vice president at Tola Capital, where he leverages a background in business planning, applying financial and metrics-based views to make strategic business and product decisions. As an investor, Jake applies this experience to do deep financial forensics to understand the underlying health and drivers of businesses and to identify strategic growth opportunities. Prior to Tola, Jake was the finance lead for the Customer Experience product line at Qualtrics where he supported both revenue and OPEX planning. Before joining Qualtrics, Jake was part of Microsoft’s Commercial Office 365 business where he supported M&A, strategic partnerships, addressable market analysis, and product roadmap planning. 

Related Items:

GenAI and the Future of Work: ‘Magic and Mayhem’

AI Is Coming for White-Collar Jobs, Too

Conversational AI Poised to Be Major Disrupter