Automation is fun; it’s my speciality. Yet, with all the AI rage, Sally from HR is wondering if her job is on the chopping block. This post addresses the upward bound of automation; in other words, what can and cannot be automated. With the increased capacity for automation via AI, it’s worth the attempt to delineated some principles by which a process can quickly be evaluated to determine its automation possibility.
I maintain that though the new tools are excellent and fun, many people now vastly over-estimate what can actually can be automated. For the purpose of this post, automation is defined as configuring technology to entirely remove work from the effort of an operator, requiring no manual intervention for executions and only occasional oversight and checks on performance. I hold that defined tasks can and should be automated, but roles performed by agents, can only be assisted.
High volume, rule governed tasks: fully automate
Any repetitive process where structured data travels through a defined funnel: this is automation gold. For example, appointment and scheduling and reminders, invoice tracking, standard requests for reviews from clients, omw texts, all-done texts, content posts to multiple platforms simultaneously, repetitive administrative communications with established relationships. Generic intake forms, repeatable data processing (a data export from one system to an import into another system). However, most of this has already been available through scripting, and AI is not bringing much new here.
What AI brings to the picture is automation of fuzzy input. Classification of bank transactions, sorting classes of incoming email, for example. However, this still requires strict and well thought through rules, and auditability. These automations, as opposed to scripts, are the most expensive to run and the most fragile in production. They cost the most and should be recommended the least. This leads naturally into the second category:
Value-based, ambiguous judgements, can be assisted.
Anything that requires value judgements, or ambiguous situations, cannot be fully automated, but can be assisted by AI. Responding to emails, composing communications (to be used with caution), cold-warm prospecting, outreach scripts, follow up sequences, building/coding, websites, or pipelines. These, I hold, can and only ever can be assisted, they are all at root expressions of the values of the operator, which must be the rooted guide to any generative AI. Any profitable production use of AI automation in these situations, still require strict scoping, monitoring, direction and wisdom, to be provided by the operator. These mentioned have left the sphere of the tasks, and have entered that of an ongoing role.
Setting up an AI agent to read and respond to all the customer emails is a great way to lose all your customers. Setting up a prospecting agent to do 1,000 cold reach-outs a day via email and text is a great way to set the world on fire. Composition aid arguably should not be used at all, as it makes the operator weaker and dumber over time, as he outsources more and more speech and intelligence. Not to mention that many people, even now, despite the claims of many token salesmen, are still perfectly capable of sensing AI vs. human written content. It’s all in the choices.
Communication and critical operation roles will always require, at minimum, the oversight of the operator. However, once an operator has determined the general concept of the response, AI can generate polished drafts with lots of em dashes — marshmallow fluff. This is ideal in situations where highly polished boiler-plate HR- approved language is most necessary, and one is happy to not waste their mind on a mind-wasting task, like responding to toll booth operators, deploying your agent to call customer service on your behalf.
That is to say, ambiguous situations which require value judgments can be assisted with automation, but only assisted. Assistance means defining values for the bot, giving iterative direction, and auditing performance, especially for anything in critical production roles.
Tasks with personal, relational, or legal accountability: never
You open up open claw and say, “do my taxes”, “run my business”, “respond to all my emails”, “text my wife”. This is a bad move. When the chickens come home to roost, you’re the one who has to sit in a judgment seat.
Any role where there is legal or relational or personal accountability, will never be able to be handed off to an autonomous agent. Men are given dominmion over the earth, men are in covenant with God, fellow man, and the state, and men are the ones who must answer for the way they spend their talents.
When Open Claw cheats on your taxes, who will uncle Sam arrest, you or your bot? When you your agent army spends all your ad money to pump slop to all channels, rendering you a negative ROI on your AI-psychosis-induced-side-hustle, who loses their assets, you or your bot? When you send Open Claw on date night with your wife, who signs the divorce papers, you or your bot?
When a man automates, he is automating toward a purpose, but his choices, his automation, his success or failure in achieving his purpose; these are required at his hand, not at the hand of his tools.
Things which shouldn’t be automated but people will still try:
Let me take a moment to be bold in three predictions:
Outbound prospecting. As people have become more accessible through technology, personal trust and accountability are the things which sell. We’re all already getting 10-15 spam/cold outbound emails and calls a day. Do you think an AI doing the calls will convert anyone worth converting? Your only “sales” will come from the nursing home.
Writing. Or at least long-form writing which quality readers will actually want to read. Who will invest time and thought into the perspective of a writer who is not capable of expressing it on his own? Who’s reading a writer who can’t write? We’re in a time of a deep dearth of soul. Your slop thesis can stay in the GPT where it belongs. We will spend our time with the men of old who had MIND!
Customer service: customers will choose the business which gives the time to talk to them. Everyone already has access to high-quality intelligence systems, and public docs and FAQ’s to attempt to solve their own problems. By the time they are reaching out to a business in desperation, a bot is the last thing they want to hear. This holds unless the Bezos-class gets their way, in which we’ll all be pushing boxes around in a warehouse until the Lord returns anyway. Maranatha!
In sum:
People are generally too materialistic in their evaluation of man, they think AI bots and humans are the same in substance, and different in form only; they both take in information, and based on deterministic if/then statements, training data, base model and memory, will process it all the same and come to deterministic results. This is false. Humans will always have a quality/substance which no organization of matter and machinery will: a soul.
A soul brings wisdom or folly, values or corruption, discernment or foolishness; a soul is the prerequisite for the establishment of trust, because it is the prerequisite for accountability, two things your sweet AI bot will never have.
Though I would be happy if we can find a way to automate Sally from HR.