You prefer a short answer when the decision is straightforward. You want options when the tradeoff matters. You care more about getting home early than finding the cheapest flight.
An assistant that picks up those details becomes easier to delegate to.
Instinct puts that idea at the center of its public description: an assistant that understands your work and priorities, reached through text or voice. The company describes tasks ranging from handling a bill to planning a trip.
The appeal is straightforward. You want to explain less because the assistant has learned what matters.
At work, that learning becomes more complicated. The person giving the instruction may be expressing a personal preference, a team convention, or a company rule. Those carry different authority.
Three instructions can carry three kinds of authority
Consider a hypothetical customer-success team.
One manager tells an AI to offer a free extension when a customer needs more evaluation time. Another says extensions must go through the account owner. Finance says an unpaid invoice changes the process.
The assistant needs to understand whose decision each instruction represents.
“Keep my customer emails under 150 words” could be the manager's personal preference. The manager can change it for their own writing.
“Include the account owner on extension requests” could be a customer-success convention. The team lead can explain when it applies and approve changes to that team's process.
“An unpaid invoice requires finance review before an extension” could be an approved company rule. A manager's preference for helping the customer does not authorize them to waive it.
An instruction can be useful without having authority over every part of the decision.
A correction can exceed the person's authority
Consider a second hypothetical example: a sales manager corrects an AI to treat a discount as approved once the buyer accepts it verbally.
The company's commercial policy requires the revenue lead's written approval before a quote includes a discount above 10 percent.
The buyer's agreement establishes interest in the offer. It does not establish the company's approval to make it. The sales manager can explain what happened in the conversation, while the revenue lead owns the discount decision.
Remembering the manager's correction as the universal quoting rule would quietly hand them authority they do not have.
A useful assistant needs the approved instruction that governs the quote, alongside the salesperson's account of the conversation.
Two roles may be talking about different problems
Take a third illustrative example.
An account director asks an AI to treat every ticket from a strategic customer as urgent. The support lead reserves Priority 1 for a production outage.
The account director is expressing the customer's commercial importance. The support lead is defining the incident's severity.
The team can preserve both meanings: flag the customer for special attention while keeping the incident's priority consistent with its severity. Changing the Priority field to satisfy every request for urgency would make the support queue less useful.
This disagreement needs the people who own those decisions to clarify the instruction. Repeating either correction more often would not resolve it.
Understanding the person makes the distinction more useful
Instinct's emphasis on understanding what a person is working on captures why personal context matters. A manager's preferred level of detail can make a briefing more useful. Their knowledge of a customer's history can improve a response.
At work, that understanding also needs to preserve which decision the person is entitled to make. The same manager may own their writing preference, help define a team convention, and need finance approval for an exception.
Our companion essay on company work's missing repository develops the broader case for giving those reviewed instructions a shared home.
HexaHQ keeps the reviewed instruction available
HexaHQ is where the company keeps reviewed knowledge, corrections, and approved workflows for use through supported AI clients. The salesperson preparing a quote can use the approved discount instruction; the support team can use the agreed meaning of Priority 1.
People still decide which instruction governs. An AI helping someone articulate a correction does not grant that person authority over another team's process.
Use your next process review to distinguish personal preferences, team conventions, and approved policy before asking AI to apply them.
The right person needs to decide before a correction becomes the team's rule.