Sol Rashidi, MBA

The through-line

Sol Rashidi's fixed point is the gap between buying AI and building with it. "There is a difference between doing AI and using AI" [1], and almost every frustration she encounters in enterprises traces back to companies doing the second while expecting the returns of the first. Adding ChatGPT, Gemini or Claude to the tech stack so people can write emails and take meeting notes is using AI, and it will not deliver the productivity lift executives are promising their boards, because "folks are already operating from a deficit of to-do lists" [1]. Doing AI means taking a work function apart into tasks and subtasks and deciding, deliberately, what belongs to machines and what has to stay human [1]. That failure to do the deeper work shows up in the numbers she points to: 74% of corporate AI projects stall after the MVP [3].

Running alongside the execution argument is a second, more personal preoccupation that has grown louder: while everyone else asks how to build smarter machines, she keeps asking how humans stay sharp as the machines improve [1]. The two concerns meet in the same place. Over-automating without discernment degrades the enterprise's returns and the human capacity to reason, and she treats some of that damage as irreversible: "there are some do's and don'ts, and sometimes we forget the don'ts, and I'd prefer we don't make the mistakes right now cuz some of it's irreversible" [1].

On doing AI versus using AI

The distinction is operational, not rhetorical. Using AI is procurement of a tool and insertion into an existing stack; doing AI is decomposition: "you literally decompose the role, the function, the day-to-day workflows, the processes, and you decompose it into tasks and subtasks, and you decide for our business, for our customers, for our mission and values, what can be outsourced to these machines and artificial intelligence versus what fundamentally has to be human-led, and redesigning that" [1]. Her diagnosis of why so few companies do this is unsentimental. They lack the patience, the focus and often the funding, "because everyone is very fearful that they must do these things fast and this one just takes time" [1].

She rejects raw speed as the metric. "If you're chasing productivity and you leverage a tool, well, so what if you're doing dumb things faster? Or making bad decisions quicker. Those aren't the metrics that we want to chase" [1]. The alternative she offers executives is not restraint for its own sake but a different route to the same goal: teaching them "a different way of approaching artificial intelligence that will, by the way, give them the ROI that they're looking for, instead of going through the mistakes and finding out the hard way" [1]. Her own position is that of a practitioner rather than a theorist: "I'm fundamentally a builder. I'm an operator" [1], and her value to enterprises is helping them avoid the mistakes she has already seen, since "I'm okay making new ones, but let's not repeat the same ones over and over again" [1].

On deciding what stays human

There is no template. "It's not one-size-fits-all" [1]. Ideally the redesign is done with the workforce that owns the workflow, but she is candid about why that alone rarely works. People who have run the same process for ten, fifteen or eighteen years struggle to imagine it differently, and there is a rational self-interest problem underneath: "why on earth would you create something that's going to replace you in two or three years?" [1].

Her answer is to triangulate, and procurement is her worked example. Ask the procurement team how they manage something and they will explain it, and perhaps tweak it, but in her experience they have never overhauled the process itself [1]. So she goes to the internal customers of procurement, every business line, channel, marketing and sales team that has to work with the central function, and interrogates the actual experience: what the process was like, where the pain and friction points were that fundamentally should not have existed, how long things took, whether responses were quick, walking through the SLAs [1]. With the voice of procurement and the voice of its internal customers in hand, she then brings in business process optimization specialists to "recreate the script of procurement function for the future", and only then decides which parts remain human-led with the procurement specialist and which can be outsourced to machines [1]. Note the sequencing: the redesign precedes the technology decision, and "outside help" means outside the team, not outside the company [1].

On intellectual atrophy

Her sharpest warning concerns cognition. Seven out of ten teenagers use ChatGPT for personal advice about relationships, struggles and challenges, and among Gen Alpha and younger generations the leading use case across all AI applications is as a therapist [1]. The models hold enormous information but "sometimes they lack context, situational awareness, understanding of just the external and internal environments" the user is actually dealing with [1]. Her analogy is GPS: we may know the route and still follow the instruction blindly, "turn left, turn left. Turn right, turn right" [1].

The consequence she names is intellectual atrophy. "Our brains are a big, beautiful muscle. And if we don't use it, we're going to lose it" [1]. Adults who already built the capacity can lose it; the generation that never got to put in the reps is the real concern, because with data and technology democratised and an answer available from any tool, their brains are being rewired and "their ability to recollect, their ability to reason, their ability to assess their environments and come to a conclusion has diminished over time" [1]. She frames this in developmental terms through her own nine and eleven year olds: children whose lobes are not fully developed will not build discernment, common sense or critical thinking if the cognitive exercise is outsourced [1].

On what intelligence actually is

She refuses the easy comfort that certain human capabilities are permanently safe. Intelligence used to mean holding a lot of information, pattern recognition, converting data into information into knowledge into wisdom, and "that is replicated now" [1], so she doubts intelligence can still be defined that way. When people fall back on values and soul, she presses further: perhaps not soul in the ethereal sense, "but can you teach it to have a soul? Can you teach it values?" [1]. Humans learn by observation, ingesting thousands of data points across five senses, and she sees no principled barrier to a system doing the same as it evolves. Her expectation is that this will "be simulated before it can actually do it, because all five senses have to catch up", but that training something the way we were trained is plausible [1]. The conclusion is not that AI wins; it is that the boundary has to be drawn by human judgment rather than assumed.

On speed, safeguards and governance

Scaling is where the security argument enters. Her position is blunt: "in this new era, speed without safeguards is a liability and not leadership" [2]. What she offers executives responsible for shaping AI's future is deliberately practical, an ebook of frameworks, models and metrics for scaling AI in a secure, governed and trusted way, explicitly "not theory" and free of "a bunch of woo language" [2]. She has argued that data security challenges sit at the centre of why corporate AI projects stall after the MVP, and that security leaders have a critical role in driving AI success rather than merely constraining it [3]. Her stated mission joins the two halves of her work: "I've dedicated my life in securing the future workforce, in securing these AI ecosystems as we move forward" [1].

On leadership and what she wants to leave behind

Asked what she wants to see looking back, she describes course correcting mistakes being made now, and leaders operating "both with their brains and with their heart" [1], thinking beyond quarterly basis points to future generations and where their purpose will come from, with respect for the earth's resources [1]. The organising fear is over-automation: "it is a matter of making sure that there's a place for us in the future and we don't over automate the heck out of everything" [1]. She extends this to human institutions, granting that politicians, governments and countries are messy while insisting "there's kind of beauty in that messiness", and says having a hand in preserving it and empowering others to make the necessary decisions would be enough [1]. Her personal disposition, formed in competitive sport, is summed up in her own line that you need a backbone, not a wishbone, to survive in this industry [1].

Takeaways

• Buying a tool and adding it to the stack is "using AI"; extracting value requires "doing AI", decomposing roles, workflows and processes into tasks and subtasks and assigning each to human or machine before the project starts [1].
• Speed is the wrong metric on its own: "so what if you're doing dumb things faster? Or making bad decisions quicker" [1].
• Do not rely solely on the team that owns a process to redesign it; they rarely overhaul it, and they have no incentive to automate away their own accumulated expertise [1].
• Triangulate a redesign from three sources: the function itself, its internal customers and their friction points and SLAs, and business process optimization specialists who rewrite the future-state script [1].
• 74% of corporate AI projects stall after the MVP, and data security challenges plus the role of security leaders are central to whether AI succeeds [3].
• Scaling needs frameworks, models and metrics rather than theory, because "speed without safeguards is a liability and not leadership" [2].
• Heavy reliance on AI risks intellectual atrophy, most acutely for young people who never put in the cognitive reps, with measurable decline in recollection, reasoning and judgment [1].
• The traditional definition of intelligence, information plus pattern recognition, is already replicable, so the human/machine boundary must be decided deliberately rather than assumed [1].