How to Treat Your New Colleague—The LLM
How to Treat Your New Colleague—The LLM
How to Treat Your New Colleague—The LLM
How to Treat Your New Colleague—The LLM
We expect magic. We get mush. Not because the capability isn’t there, but because we’ve treated our LLM like a mind reader instead of a colleague who happens to know a lot, but knows nothing about your business until you tell it.
Share
We expect magic. We get mush. Not because the capability isn’t there, but because we’ve treated our LLM like a mind reader instead of a colleague who happens to know a lot, but knows nothing about your business until you tell it.
Share
We expect magic. We get mush. Not because the capability isn’t there, but because we’ve treated our LLM like a mind reader instead of a colleague who happens to know a lot, but knows nothing about your business until you tell it.
I’ve spent the last few months on an outrageous number of customer calls, working with teams as they try to scope, test, and implement AI into their day-to-day. The ambition and capability is there—but one theme keeps surfacing: most people struggle with how to instruct AI to complete the task. In other words: prompting.
It sounds trivial, but it’s not. Get it wrong and your LLM gives you a polite muddle and you lose hope. Get it right and it becomes a trusted colleague you can actually rely on. What follows isn’t a grand thesis on AI, but some tactical reflections from the field—patterns I’ve seen, mistakes worth avoiding, and practices that help teams get started faster and get value sooner.
I’ve spent the last few months on an outrageous number of customer calls, working with teams as they try to scope, test, and implement AI into their day-to-day. The ambition and capability is there—but one theme keeps surfacing: most people struggle with how to instruct AI to complete the task. In other words: prompting.
It sounds trivial, but it’s not. Get it wrong and your LLM gives you a polite muddle and you lose hope. Get it right and it becomes a trusted colleague you can actually rely on. What follows isn’t a grand thesis on AI, but some tactical reflections from the field—patterns I’ve seen, mistakes worth avoiding, and practices that help teams get started faster and get value sooner.
I’ve spent the last few months on an outrageous number of customer calls, working with teams as they try to scope, test, and implement AI into their day-to-day. The ambition and capability is there—but one theme keeps surfacing: most people struggle with how to instruct AI to complete the task. In other words: prompting.
It sounds trivial, but it’s not. Get it wrong and your LLM gives you a polite muddle and you lose hope. Get it right and it becomes a trusted colleague you can actually rely on. What follows isn’t a grand thesis on AI, but some tactical reflections from the field—patterns I’ve seen, mistakes worth avoiding, and practices that help teams get started faster and get value sooner.
The First Day on the Job
A scenario everyone is familiar with: a new colleague joins your team. Bright-eyed, fresh from a glowing interview process, capable of brilliance. You give them their laptop, point in the general direction of the backlog, and mutter something like, “Go fix things.”
No role, no context, no data, no instructions. Just expectations.
A week later, you’re scratching your head, wondering why nothing useful has landed. This, in short, is how most people treat an LLM.
We expect magic. We get mush. We expect Shakespeare. We get fridge poetry. Not because the capability isn’t there, but because we’ve treated it like a mind reader instead of what it really is: a colleague who happens to know a lot, but knows nothing about your business until you tell it.
And so the lesson is this: if you want quality output, you must treat your LLM like you would a newly hired colleague.
“We expect magic. We get mush. We expect Shakespeare. We get fridge poetry. Not because the capability isn’t there, but because we’ve treated [our LLM] like a mind reader instead of what it really is: a colleague who happens to know a lot, but knows nothing about your business until you tell it.”
Julia Sommarlund
Product Manager
Planhat
The First Day on the Job
A scenario everyone is familiar with: a new colleague joins your team. Bright-eyed, fresh from a glowing interview process, capable of brilliance. You give them their laptop, point in the general direction of the backlog, and mutter something like, “Go fix things.”
No role, no context, no data, no instructions. Just expectations.
A week later, you’re scratching your head, wondering why nothing useful has landed. This, in short, is how most people treat an LLM.
We expect magic. We get mush. We expect Shakespeare. We get fridge poetry. Not because the capability isn’t there, but because we’ve treated it like a mind reader instead of what it really is: a colleague who happens to know a lot, but knows nothing about your business until you tell it.
And so the lesson is this: if you want quality output, you must treat your LLM like you would a newly hired colleague.
“We expect magic. We get mush. We expect Shakespeare. We get fridge poetry. Not because the capability isn’t there, but because we’ve treated [our LLM] like a mind reader instead of what it really is: a colleague who happens to know a lot, but knows nothing about your business until you tell it.”
Julia Sommarlund
Product Manager
Planhat
The First Day on the Job
A scenario everyone is familiar with: a new colleague joins your team. Bright-eyed, fresh from a glowing interview process, capable of brilliance. You give them their laptop, point in the general direction of the backlog, and mutter something like, “Go fix things.”
No role, no context, no data, no instructions. Just expectations.
A week later, you’re scratching your head, wondering why nothing useful has landed. This, in short, is how most people treat an LLM.
We expect magic. We get mush. We expect Shakespeare. We get fridge poetry. Not because the capability isn’t there, but because we’ve treated it like a mind reader instead of what it really is: a colleague who happens to know a lot, but knows nothing about your business until you tell it.
And so the lesson is this: if you want quality output, you must treat your LLM like you would a newly hired colleague.
“We expect magic. We get mush. We expect Shakespeare. We get fridge poetry. Not because the capability isn’t there, but because we’ve treated [our LLM] like a mind reader instead of what it really is: a colleague who happens to know a lot, but knows nothing about your business until you tell it.”
Julia Sommarlund
Product Manager
Planhat
Give it a Role, Give it a Goal
When you onboard a new team member, you don’t just say “make yourself useful.” You say, “You’re our customer success manager. Your goal is to reduce churn. Here’s how we measure you. This is what success looks like, this is what failure looks like. This is what you should do if you can’t complete the task.”
Your LLM is no different. Without roles or objectives, it will default to bland generalities. Assign it a role—researcher, analyst, sales rep, technical writer. Tell it the goal—persuade, summarise, compare, calculate.
Think of it like setting the sat-nav: you need both a starting point and a destination. Without both, you’re just circling the roundabout—endlessly polite, but going nowhere.
Give it a Role, Give it a Goal
When you onboard a new team member, you don’t just say “make yourself useful.” You say, “You’re our customer success manager. Your goal is to reduce churn. Here’s how we measure you. This is what success looks like, this is what failure looks like. This is what you should do if you can’t complete the task.”
Your LLM is no different. Without roles or objectives, it will default to bland generalities. Assign it a role—researcher, analyst, sales rep, technical writer. Tell it the goal—persuade, summarise, compare, calculate.
Think of it like setting the sat-nav: you need both a starting point and a destination. Without both, you’re just circling the roundabout—endlessly polite, but going nowhere.
Give it a Role, Give it a Goal
When you onboard a new team member, you don’t just say “make yourself useful.” You say, “You’re our customer success manager. Your goal is to reduce churn. Here’s how we measure you. This is what success looks like, this is what failure looks like. This is what you should do if you can’t complete the task.”
Your LLM is no different. Without roles or objectives, it will default to bland generalities. Assign it a role—researcher, analyst, sales rep, technical writer. Tell it the goal—persuade, summarise, compare, calculate.
Think of it like setting the sat-nav: you need both a starting point and a destination. Without both, you’re just circling the roundabout—endlessly polite, but going nowhere.
Feed it Context, Not Crumbs
A colleague needs background. Customer and contact history, market data, team notes. Deprive them of this, and they’ll make educated guesses—emphasis on guesses.
So too with an LLM. It thrives on context. Without it, you’ll get the equivalent of pub chatter: entertaining, sometimes insightful, but hardly the stuff of board decks. With it, you’ll get something closer to well-prepared consultancy.
Context is the difference between:
“Write me an email to a prospect.” Generic, soulless, could be for anyone.
“Write me an email to prospect X. I want to sell a CRM.” Actionable, specific, yours.
Or even better, include real-time data from your CRM, and skip the work of adding manual input yourself by providing rich context for the LLM to deeply understand the situation:
“Write an email to prospect X. My contact is the CEO of a software company, here are the last conversations we’ve had with them, and this is the sentiment for each conversations. Here are the key stakeholders, here’s details on the opportunity.” Specific, anchored in context, including every data point that could be of interest.
Feed it crumbs and you’ll get crumbs back. Feed it the meal, and it will set the table.
Feed it Context, Not Crumbs
A colleague needs background. Customer and contact history, market data, team notes. Deprive them of this, and they’ll make educated guesses—emphasis on guesses.
So too with an LLM. It thrives on context. Without it, you’ll get the equivalent of pub chatter: entertaining, sometimes insightful, but hardly the stuff of board decks. With it, you’ll get something closer to well-prepared consultancy.
Context is the difference between:
“Write me an email to a prospect.” Generic, soulless, could be for anyone.
“Write me an email to prospect X. I want to sell a CRM.” Actionable, specific, yours.
Or even better, include real-time data from your CRM, and skip the work of adding manual input yourself by providing rich context for the LLM to deeply understand the situation:
“Write an email to prospect X. My contact is the CEO of a software company, here are the last conversations we’ve had with them, and this is the sentiment for each conversations. Here are the key stakeholders, here’s details on the opportunity.” Specific, anchored in context, including every data point that could be of interest.
Feed it crumbs and you’ll get crumbs back. Feed it the meal, and it will set the table.
Feed it Context, Not Crumbs
A colleague needs background. Customer and contact history, market data, team notes. Deprive them of this, and they’ll make educated guesses—emphasis on guesses.
So too with an LLM. It thrives on context. Without it, you’ll get the equivalent of pub chatter: entertaining, sometimes insightful, but hardly the stuff of board decks. With it, you’ll get something closer to well-prepared consultancy.
Context is the difference between:
“Write me an email to a prospect.” Generic, soulless, could be for anyone.
“Write me an email to prospect X. I want to sell a CRM.” Actionable, specific, yours.
Or even better, include real-time data from your CRM, and skip the work of adding manual input yourself by providing rich context for the LLM to deeply understand the situation:
“Write an email to prospect X. My contact is the CEO of a software company, here are the last conversations we’ve had with them, and this is the sentiment for each conversations. Here are the key stakeholders, here’s details on the opportunity.” Specific, anchored in context, including every data point that could be of interest.
Feed it crumbs and you’ll get crumbs back. Feed it the meal, and it will set the table.
Be Precise, Then Iterate
Good managers don’t throw tasks over the fence. They brief, check, course-correct. They say “try this,” then refine.
Working with an LLM is no different. The clearer your instructions, the better the first draft. Then, treat it like an iterative colleague: review, adjust, and ask again.
Specificity isn’t pedantry, it’s productivity. “Give me a three-paragraph summary highlighting risks for the executive team, using bullet points for financial metrics” will beat “summarise this” every day of the week.
And don’t be shy to say, “That’s not quite it, try again but with a different tone.” Iteration is not inefficiency; it’s collaboration..
“Your LLM is not a seer. It’s an energetic, full-of-potential colleague. Treat it accordingly.”
Julia Sommarlund
Product Manager
Planhat
Be Precise, Then Iterate
Good managers don’t throw tasks over the fence. They brief, check, course-correct. They say “try this,” then refine.
Working with an LLM is no different. The clearer your instructions, the better the first draft. Then, treat it like an iterative colleague: review, adjust, and ask again.
Specificity isn’t pedantry, it’s productivity. “Give me a three-paragraph summary highlighting risks for the executive team, using bullet points for financial metrics” will beat “summarise this” every day of the week.
And don’t be shy to say, “That’s not quite it, try again but with a different tone.” Iteration is not inefficiency; it’s collaboration..
“Your LLM is not a seer. It’s an energetic, full-of-potential colleague. Treat it accordingly.”
Julia Sommarlund
Product Manager
Planhat
Be Precise, Then Iterate
Good managers don’t throw tasks over the fence. They brief, check, course-correct. They say “try this,” then refine.
Working with an LLM is no different. The clearer your instructions, the better the first draft. Then, treat it like an iterative colleague: review, adjust, and ask again.
Specificity isn’t pedantry, it’s productivity. “Give me a three-paragraph summary highlighting risks for the executive team, using bullet points for financial metrics” will beat “summarise this” every day of the week.
And don’t be shy to say, “That’s not quite it, try again but with a different tone.” Iteration is not inefficiency; it’s collaboration..
“Your LLM is not a seer. It’s an energetic, full-of-potential colleague. Treat it accordingly.”
Julia Sommarlund
Product Manager
Planhat
The Myth of the Magic Box
There’s a myth that AI is plug-and-play, a sort of enchanted typewriter that turns vague ideas into polished brilliance. The reality is more mundane—and more powerful.
It is a systems problem. Like humans, an agent needs context, tools, and a clear sense of trade-offs. Without those, you end up with fluff. With them, you end up with function.
And so the art is not in “prompting,”—that much-abused word—but in managing. The best results come not from incantations but from the same principles that make teams work: clarity, context, iteration.
Your LLM is not a seer. It’s an energetic, full-of-potential colleague. Treat it accordingly:
Set the role and objective
Provide the context
Give clear instructions, then refine
Do this, and you’ll move from outputs to outcomes. You’ll spend less time rolling your eyes at generic summaries, and more time unlocking genuine leverage. The truth is, AI doesn’t make people redundant. It makes bad habits redundant. The habit of vague asks, the habit of under-preparing, the habit of expecting brilliance without briefing.
Treat your LLM as you would a promising new hire. Respect it, guide it, and it will surprise you. Neglect it, and it will muddle along.
Either way, you get the colleague you deserve.
The Myth of the Magic Box
There’s a myth that AI is plug-and-play, a sort of enchanted typewriter that turns vague ideas into polished brilliance. The reality is more mundane—and more powerful.
It is a systems problem. Like humans, an agent needs context, tools, and a clear sense of trade-offs. Without those, you end up with fluff. With them, you end up with function.
And so the art is not in “prompting,”—that much-abused word—but in managing. The best results come not from incantations but from the same principles that make teams work: clarity, context, iteration.
Your LLM is not a seer. It’s an energetic, full-of-potential colleague. Treat it accordingly:
Set the role and objective
Provide the context
Give clear instructions, then refine
Do this, and you’ll move from outputs to outcomes. You’ll spend less time rolling your eyes at generic summaries, and more time unlocking genuine leverage. The truth is, AI doesn’t make people redundant. It makes bad habits redundant. The habit of vague asks, the habit of under-preparing, the habit of expecting brilliance without briefing.
Treat your LLM as you would a promising new hire. Respect it, guide it, and it will surprise you. Neglect it, and it will muddle along.
Either way, you get the colleague you deserve.
The Myth of the Magic Box
There’s a myth that AI is plug-and-play, a sort of enchanted typewriter that turns vague ideas into polished brilliance. The reality is more mundane—and more powerful.
It is a systems problem. Like humans, an agent needs context, tools, and a clear sense of trade-offs. Without those, you end up with fluff. With them, you end up with function.
And so the art is not in “prompting,”—that much-abused word—but in managing. The best results come not from incantations but from the same principles that make teams work: clarity, context, iteration.
Your LLM is not a seer. It’s an energetic, full-of-potential colleague. Treat it accordingly:
Set the role and objective
Provide the context
Give clear instructions, then refine
Do this, and you’ll move from outputs to outcomes. You’ll spend less time rolling your eyes at generic summaries, and more time unlocking genuine leverage. The truth is, AI doesn’t make people redundant. It makes bad habits redundant. The habit of vague asks, the habit of under-preparing, the habit of expecting brilliance without briefing.
Treat your LLM as you would a promising new hire. Respect it, guide it, and it will surprise you. Neglect it, and it will muddle along.
Either way, you get the colleague you deserve.
Planhat AI Platform
Planhat AI Platform
Planhat AI Platform
Customers
© 2025 Planhat AB
Customers
© 2025 Planhat AB
Customers
© 2025 Planhat AB
Customers
© 2025 Planhat AB