Business
What a consultant delivers when agents do the work
A consultant whose agents do the work sells an outcome, the accountability for that outcome arriving, and the right to stop the system. Hours measure the work that is cheapest to remove.
A consultant whose agents do the work sells the accountability for an outcome and the right to stop it. The hour has stopped being a useful unit, because hours measure the work that is cheapest to remove. What the customer still has to pay for is who answers when the system gets it wrong.
Apps AS has delivered consulting work since 2008 and built its own products alongside it. This article is based on changes we have made in our own business, not on estimates of what the industry is going to do.
What does a consultant sell when agents do the work?
A consultant sells three things when agents do the work: an outcome defined in advance, accountability for that outcome arriving, and a named person who can stop the system. None of the three is measured in hours.
An outcome is a result the customer recognises without looking inside the system: the case is closed, the invoice is credited, the meeting is booked. An agent in production is not an outcome. Neither is a report.
Accountability means the supplier pays when the outcome fails to arrive. If there is no consequence for the supplier, the accountability is a sentence in a proposal and not a guarantee.
The right to stop is the part customers most often forget to buy. An agentic system nobody has the authority to switch off is a risk the customer has taken on without having it priced.
What went first in our own trade?
The pre-project went first. Requirements gathering, user stories, a functional specification and an estimate make up the part of consulting work that is easiest to describe, and therefore the part an agent does first.
Apps AS built Adviseful.ai for precisely that job. The tool takes a customer from idea to a fully specified app concept in minutes: AI-generated concepts, user stories, a summary with estimates for development time and resource use, and a PowerPoint the customer can take away.
Two things about that decision are worth noticing. Apps AS automated its own billable phase, not the customer's. And Apps AS sells the tool on to IT consultancies and digital agencies, which is to say to companies that make their living from the same phase.
Apps AS made a related move once before, with AppCloud. The customer stopped paying for bespoke work and subscribed instead to design, development and operations at a fixed monthly price. The price moved from hours to delivery before agents arrived.
What is left for the human?
What is left for the human is the decisions that have no field in any form. Three of them have a place in the delivery and a price, and each is worth writing into the contract on its own.
- Before delivery
The list of actions that will not be automated. Someone goes through what the system is allowed to do on its own and what requires an approval. The list is short, it is known in advance, and it is expensive to produce afterwards.
The list sets the price. If the action can be rolled back, the supplier can guarantee an outcome, because the error can be corrected. If the action cannot be rolled back, the supplier is selling an approval surface and an accountability, and the price should say so. How a technical choice ties to a price tag is covered in Architecture, design and price are the same decision.
- During delivery
The judgement that has no field. An agentic system delivers work that passes every machine check and is still wrong for reasons nobody has built a field for. Someone has to read the result and recognise that kind of error.
Our own publishing agent on apps.no is the example closest to hand. On 21 September 2026 it delivered two complete article drafts that passed schema validation in the CMS. A human rejected both: one because the evidence was second-hand, the other because the numbers belonged to a different company than the blog belongs to. Neither reason has a field. What that means for running agentic systems is covered in What changes when software acts on its own.
That judgement is what the consultant actually gets paid for, and it takes minutes. Pricing by the hour penalises whoever spends five minutes rather than five hours making it.
- After handover
Ownership of the process. When the system keeps running without the supplier, one named person at the customer has to own the target metric and hold the authority to stop. Without that name, the handover is simply the end of the engagement.
Who owns the process after automation is the same question seen from the business's side, and it is covered in Personalization is an operations task, not a campaign.
What can we promise, and what can we not promise?
We can promise an outcome where the action can be rolled back, and we can promise a response time for correcting it. We cannot promise that an agentic system will never get anything wrong. A proposal that promises that was written by someone who has not run one in production.
Staffing changes accordingly. A team that delivers outcomes needs fewer people producing first drafts and more people who can decide whether a draft holds up. Those skills are harder to hire than the ones they replace, because they assume the person has done the work themselves long enough to recognise when it is wrong.
Logging and traceability point the same way. The log is both the debugging trail and the documentation a public-sector buyer will ask for. Why that demand arrives regardless of whether the regulation has landed yet is set out, from the Norwegian case, in Agentic AI in a country with no AI law.
Write down what you are accountable for
If you are selling an agentic system to a customer this year, write three lines before the proposal goes out: the outcome you guarantee, the actions you will not automate, and the name of the person at the customer who can stop the system. Send those three lines to hello@apps.no. We will reply with which of them we would price highest, and which one we would cut.
Sources
About this article. Written by Kjell Rune Monsø, co-founder and partner at Apps. Anchor article for the pillar The future of consultants. The details about Adviseful.ai and AppCloud come from Apps AS's own product pages, read 24 September 2026. The publishing agent example is based on the runs of 21 September 2026. How many consultancies use Adviseful.ai is not stated on the product page and is not estimated here. The article uses no customer figures.
FAQ
- How do we price an outcome when we do not know how many cases will arrive?
- Set a price per outcome and a cap on the number within the period, so the volume risk is shared. The customer pays for what was actually delivered, and you are protected against a peak you do not have the capacity for. Make the first period short enough that both sides can adjust the price once the real volume is known, ideally three months.
- What do we do with the people who bill for pre-project work today?
- Most of them can make the judgement the tool does not, which is deciding whether a draft holds up and what is missing. The transition is real and uncomfortable, because the same person delivers fewer billable hours and more value per hour. It is a pricing problem before it is a staffing problem, and companies that take it in the opposite order lose people they needed.
- Can we charge for operating a system we did not build ourselves?
- Yes, and it is often easier to sell than building one. The condition is that you have access to the log and the authority to stop the system. Without both, you are accepting accountability for something you cannot influence, and then the price is too low whatever it is.
- How large does the accountability have to be before a customer believes it?
- Large enough to show up in your accounts, and small enough that one bad quarter does not sink the company. In practice many land on putting a share of the fee at stake each period. What matters is not the size but that the consequence is written down and that the customer does not have to negotiate for it when it happens.
- What happens to the price if the customer buys the model themselves?
- The price of the model falls towards zero, and that is not where the margin sits anyway. You are selling the integration with the customer's own systems, the list of actions that will not be automated, and the accountability for the outcome. None of those three comes with buying access to a model.
