I have been interested in automation for much longer than the current AI cycle.
When I first read Tim Ferriss’s The 4-Hour Workweek, the important idea for me was not the literal promise of working four hours. It was the challenge to stop treating every repeated task as something that had to be done manually, in the same way, forever.
Since then, I have applied that thinking to sales processes, email templates, frequently asked questions, routine replies, reporting and data consolidation. The objective was always the same: remove unnecessary repetition so more time remains for customers, decisions and growth.
Generative AI and agents change the scale of that opportunity. A knowledge worker can now research more widely, prototype tools, consolidate information and prepare work at a speed that previously required a larger team. But I do not believe the answer is to automate everything.
The more useful question is:
Which work should AI perform for me, which work should we do together, and which work must remain mine?
That is the shift I am making from automation to AI deputisation.
Domain judgement is the foundation
Two recent episodes of The AI Daily Brief gave useful language to something I have been learning through practice.
In The AI Engineering Skills Map for Knowledge Workers, Nathaniel Whittemore describes five increasingly important capabilities:
- Mapping what AI can and cannot do reliably
- Giving it the right context, tools and operating environment
- Prototyping solutions rather than only performing tasks manually
- Identifying opportunities that were previously too expensive or time-consuming
- Learning new skills continuously as the tools change
Underneath all five sits domain judgement: the ability to define quality, recognise trade-offs, understand consequences and take responsibility for the decision.
This is the part of the AI discussion that matters most to me.
AI can prepare a competitor comparison, but it does not automatically know which difference will matter to a distributor, retailer, property manager or software partner. It can draft an outreach message, but it does not know whether using a particular relationship is appropriate. It can surface a trend in a dashboard, but it cannot carry responsibility for the commercial decision that follows.
Experience becomes more valuable when AI increases the amount of information and output an operator can produce. The advantage does not come from generating more material. It comes from knowing what deserves attention, what is weak, what is missing and what to do next.
Context is more important than a clever prompt
An AI model may be capable of performing a task and still produce poor work because it lacks the context surrounding that task.
A generic prompt does not contain a company’s history, product constraints, pricing, customers, previous decisions, evidence standards or current priorities. Without that information, the output may sound polished while being commercially wrong.
This is why I built the Frankow AI Brain as a structured operating system rather than a collection of prompts.
It gives AI access to the relevant business context: workspace files, goals, product information, research, evidence, open decisions, previous discussions and rules about what can be treated as fact. The AI Board then uses different commercial perspectives to challenge a decision, while the decision history prevents useful learning from disappearing after one conversation.
The system does not make the final decision. It improves the preparation around it.
This distinction matters. Good AI work depends on both the model and the environment built around it: instructions, examples, source material, tools, permissions, memory and review rules. If that environment is weak, a more powerful model only produces weak work faster.
From automation to AI deputisation
Traditional automation is often designed to remove a task from human attention. A fixed trigger produces a fixed action: move data, send a notification, update a field or generate a standard reply.
That remains useful. But much of knowledge work is not fixed enough to disappear into a simple workflow.
The AI Deputization Audit proposes a better way to decide what AI should take on. It assesses recurring work across five practical questions:
- Does the task happen often enough, and take enough time, to justify handing it over?
- Can the process be demonstrated or taught clearly?
- Can the output be checked much faster than producing it manually?
- What happens if the AI gets it wrong?
- Does the quality depend on me personally doing the work?
The framework separates work into three broad groups: tasks that can be deputised, tasks where the human and AI work as a duet, and tasks the operator should defend and retain.
That reflects how I increasingly use AI.
Deputise the repeatable preparation
AI can take a larger role when the task is repetitive, the rules are clear, the result is easy to verify and a mistake is reversible.
Examples in my work include:
- Consolidating information from several source files
- Formatting research into a consistent structure
- Checking whether required sections or fields are missing
- Preparing first-pass company profiles
- Updating a tracking dashboard from approved data
- Turning meeting notes into actions and open questions
- Producing a first draft from an agreed brief
I do not need to personally move every fact into the right heading. I need to make sure the source is credible, the structure is useful and the conclusion is not overstated.
Work in duet where judgement shapes the output
Most valuable knowledge work currently sits in the middle.
Competitor research is a good example. AI can search more widely, organise features, compare pricing structures and identify apparent patterns. I still need to decide whether two companies genuinely compete, which difference matters commercially and whether the available evidence supports the conclusion.
The same applies to account selection and outreach. My Outreach Engine can support prospect research, qualification, message preparation and follow-up tracking. It allows a more relevant approach than sending one generic email to a large database.
But the system should not decide, without review, which relationship to use, what promise to make or whether a message is ready to send. The useful model is AI-assisted preparation followed by operator approval.
This is also how I approach commercial proposals, route-to-market comparisons and the first layer of dashboard analysis. AI expands the work I can examine. I remain involved where interpretation changes the decision.
Defend the work that depends on trust and accountability
Some work should stay human-led.
I retain direct control over:
- Final commercial recommendations
- Sensitive relationship decisions and introductions
- Negotiation positions and commitments
- Approval of claims presented as facts
- High-stakes or irreversible actions
- Messages where the recipient reasonably expects to hear from me
AI can help me prepare for a negotiation. It should not negotiate an important relationship in my name without explicit control. It can challenge my recommendation. It should not carry accountability for the outcome.
The boundary may change as the tools improve, but technical capability is not the only consideration. Privacy, security, trust, judgement and the cost of a mistake remain part of the decision.
How I apply this to market research and business development
My commercial work combines several activities that benefit from AI without becoming fully automated.
Building a better competitor view
A normal competitor table can become outdated quickly and often reduces a market to visible features.
I use AI-assisted research to collect and structure product positioning, pricing, integrations, target customers, partnerships and routes to market. The system can process a broader set of sources and highlight conflicts or gaps.
My role is to turn that material into a point of view: where the offer is genuinely differentiated, which customer has the strongest reason to change and which competitors or alternatives matter in that situation.
The research becomes useful when it changes the commercial focus, not when it produces the longest spreadsheet.
Creating targeted outreach rather than automated noise
AI makes it easy to generate high volumes of sales messages. That is not the opportunity I find most interesting.
The better use is to research a smaller number of relevant organisations, understand why each may care and prepare a message connected to its actual business. The Outreach Engine helps organise this process and preserve the evidence behind an approach.
I explain the wider method in why I research the market before making European sales introductions. The objective is not to remove the human relationship. It is to arrive at the conversation better prepared.
Building dashboards that reduce reporting work
Knowledge workers have traditionally spent substantial time collecting data before they can interpret it.
AI-assisted tools now make it possible to build lightweight dashboards that consolidate approved information, surface changes and prepare a first analytical layer. This can include account progress, competitor movements, experiment results, channel activity and open decisions.
The dashboard does not replace the commercial review. It reduces the time spent assembling the picture so more time can be used to ask why something changed and what response is appropriate.
This is one of the most practical ways AI allows a small operation to behave with the information discipline of a larger team.
The real advantage belongs to people willing to learn
The tools are changing too quickly for AI capability to become a one-off qualification.
Useful adoption requires a willingness to experiment, make mistakes, compare approaches and rebuild parts of the workflow when a better method appears. That can feel inefficient in the short term, especially for experienced people who already have a reliable way of working.
I have found the opposite over time. The willingness to learn a new tool, build a small prototype or document a process creates cumulative leverage. One experiment may save only a few minutes. A connected system of research, context, outreach and tracking changes the capacity of the whole operation.
This does not mean adopting every new model or agent. Capability mapping includes knowing when the existing process is already good enough.
The strongest knowledge workers will not be the people who automate the most. They will be the people who learn where AI adds value, build the context it needs, check its work intelligently and keep responsibility for the decisions that matter.
My operating rule
The principle I now use is simple:
Deputise the repeatable work. Work in duet where AI increases range. Defend the decisions that depend on judgement, trust or accountability.
That approach is more realistic than trying to automate an entire role. It also creates a better path for a small consultancy or commercial team to scale without pretending that every part of the work is interchangeable.
My AI-assisted Amazon and marketplace work and the OffLead.Space operating model show two different applications: using structured analysis to improve commercial decisions, and combining connected access, software and AI-assisted outreach around a new venture.
The systems will continue to evolve. The objective remains consistent: use technology to expand useful work, preserve learning and make better commercial decisions—not to remove the operator from work where the operator is the reason it has value.
Frequently asked questions
What is AI deputisation?
AI deputisation means giving an AI system a defined part of a job to perform within clear context, rules and approval boundaries. Unlike traditional automation, the operator still reviews the work, handles exceptions and remains accountable for the result.
Which knowledge-work tasks are best suited to AI?
The strongest candidates are recurring and time-consuming tasks that can be taught clearly, checked faster than they can be produced, corrected without serious consequences and completed without requiring the operator's personal judgement or relationship.
Why does domain knowledge still matter when using AI?
Domain knowledge helps an operator define quality, recognise weak assumptions, understand trade-offs and turn a plausible AI output into an appropriate commercial decision. AI can increase production capacity, but it does not remove responsibility for judgement.
How do I use AI in commercial market development?
I use AI to organise research, compare competitors, structure evidence, prepare personalised outreach, maintain decision history and consolidate tracking data. I retain control over positioning, prioritisation, sensitive communications, negotiations and final recommendations.
