This is one of the questions I am currently working on through L&F CG.
For years, I have built and rebuilt commercial playbooks around the same practical problem: how do you understand the right market, notice the right change, understand the people involved, choose the right commercial move and keep learning after the first sale?
The interesting question now is no longer only whether AI can help with individual sales tasks. It is whether the commercial process itself can be designed as a learning system that gradually needs less human administration while becoming better at knowing when a human actually adds value.
Where the question came from
Across earlier L&F CG sales plans and playbooks, a recurring logic appears:
market understanding → observation → signal → hypothesis → validation → buying process → commercial timeline → first mandate → delivery → account growth
The tools and some of the tactics have changed. The underlying mechanism has not changed as much.
Research creates hypotheses. Conversations should verify, reject or update them. A real event in the customer's world should shape timing. The first engagement should be the smallest useful next step rather than an attempt to sell everything. Delivery should create new knowledge about the account.
That looks less like a traditional sales funnel and more like a commercial decision and learning loop.
Current working idea
The system should separate what it has observed from what it merely believes.
An observed management change is evidence. "This company is now ready to buy" is a hypothesis. The system should be able to hold both ideas without confusing them.
A possible operating loop is:
market → observation → signal → hypothesis → counter-evidence → qualification → commercial motion → interaction → discovery → verified buying reality → commercial timeline → decision → outcome → learning
Sometimes the correct next action should be contact. Sometimes it should be research. Sometimes it should be to wait for a specific event. Sometimes the correct answer should simply be: we do not know enough yet.
The autonomy question
I do not want to automate a static playbook and call that intelligence.
The more useful ambition is for each part of the process to move, when evidence supports it, through something like:
human execution → human approval → guarded autonomy → autonomous execution with sampling
Research, monitoring, enrichment, meeting preparation and internal account updates may become highly autonomous quite quickly. Pricing, consequential commitments and important relationship decisions may stay human-controlled much longer.
The important part is that human corrections become learning data rather than disappearing into the next meeting or email.
If a person repeatedly changes the system's recommendation, the useful question is not only "what did the human choose?" but why was the system wrong, in what context, and does the same pattern appear again?
Why this may matter
Most commercial systems store activity. Many automate activity. Far fewer appear designed around evidence, uncertainty, learning and declining dependence on human administration.
If this works, the goal is not a sales robot. The goal is a commercial system that can increasingly handle the repetitive cognitive and administrative work itself, while escalating the moments where judgment, trust, creativity, negotiation or authority genuinely matter.
It should also be possible to expose a safe, structured part of the same commercial knowledge to the market. As buyers increasingly use their own AI agents to research and qualify suppliers, L&F CG should be understandable not only to people but also to the systems acting on their behalf.
What I still do not know
How quickly can different commercial activities safely move toward autonomy? What evidence threshold should be required before a human approval can be removed? How much learning can generalize between companies, people and markets without losing context? How should an agent distinguish a useful buying signal from an interesting but irrelevant event? And how should commercial knowledge be exposed to buyer agents without exposing private account intelligence or internal reasoning?
Those are the questions I am working through before writing the production code.