Once an AI knows your project plan, it answers many questions more precisely than you probably expect. Some decisions, however, it deliberately leaves to you, and for good reason: your project plan does not capture every connection and constraint of your project. Only you know those.
The video for this article
The video (in German) is embedded in the German version of this article.
Question 1: Everything is in the plan
The first question for the AI is about dependencies and dates in the project plan:
Prompt: Which tasks are blocking the milestone?
Which tasks in the project {PROJECT} must be completed before the milestone {MILESTONE} can be reached, and which of them are behind schedule?Open this prompt in the library ›
Via the MCP server, the AI retrieves the dependencies and the schedule status from Merlin Project and returns a list of the affected tasks along with their delay. In the test, the tasks it named matched the outline. Without MCP, you would need several views or a filter for this.
The AI answers such questions reliably because every piece of information it needs sits in the plan as a value: which task depends on which, which date is planned, which one has been reached.
Question 2: The answer lies outside the project plan
The second question sounds just as concrete:
Prompt
When will I receive the building permit so I can start the project on {DATE}?Right at the start, the AI says openly that it cannot derive the answer from the plan because the necessary information is missing from it. Instead, it weighs things up based on the available data, makes assumptions, and suggests a few options. It gives no binding answer, and that is right. Whether you will get the building permit, what you discussed with the staff at the authority, where things stand with the authority right now, or what you agreed verbally in the last meeting: none of that is in any project plan.
If an AI does give a clear recommendation on such questions, take a look at its reasoning. In most cases it then relies only on costs and dates from the plan and lacks the rest of the picture. This is where every AI reaches its limits, because your knowledge as a project manager is not always in the project plan.
The rule behind it
| In the plan | Not in the plan |
|---|---|
| Status and delay | Verbal agreements |
| Dependencies and critical path | Priorities set by management |
| Dates, durations, assignments | Mood and workload in the team |
| Summaries and reports | Strategic goals outside the project |
| Notes and attachments | External information and decisions |
The AI answers questions from the left column directly from the data. For questions that require knowledge from the right column, it weighs things up. The left column has a rule too: before you put a number into an email to management, check it briefly in the plan. AI assistants can misread values or assign them to the wrong item.
How to improve the assessment
Whatever is not in the plan, you simply write into the prompt. Add one sentence to the question, for example:
“According to the building authority, the building permit is being processed and will probably arrive next week.”
The AI factors this information into its assessment and orders the options accordingly.
The prompt library has a version that asks for this context right away:
Prompt: What are my options if the building permit is delayed?
What are my options if the building permit is delayed and the construction start on {DATE} cannot be held? Current status at the building authority: {PERMIT_STATUS}. Name the tasks that can be brought forward without the permit and rate each option by its impact on the end date.Open this prompt in the library ›
The article Prompt basics for project managers explains the principle behind this in more detail: give the AI the context it cannot find on its own.
A second limit: the AI changes nothing
The Merlin Project MCP server only reads. The AI tells you what happens if a milestone moves. Moving it is up to you. How this read-only approach protects your data is explained in the article MCP and data protection.
Try it yourself
You can test both prompts with the sample project “Single-Family Home”, which Merlin Project includes as a template. Or ask the AI your own questions. You find ideas for them here:
What do you want to know about your project?
Pick a topic. You get matching prompts that you can copy and give to your AI as soon as it is connected to Merlin Project via MCP.
Pick a topic above to see matching prompts.
Status
3 prompts to copy
What is the current status of my project? Summarize progress, delays, and open tasks in five sentences.
The AI answers from the planWhich tasks are behind schedule, and which milestones does that put at risk?
The AI answers from the planWhich tasks should already have started according to the plan but show no progress yet?
The AI answers from the planDates
3 prompts to copy
Which tasks are on the critical path?
The AI answers from the planWhich milestones are due in the next 30 days?
The AI answers from the planWhat happens to the finish date if the task {TASK} slips by {NUMBER} weeks?
The AI answers from the planResources
3 prompts to copy
Which resources are overloaded in the next four weeks?
The AI answers from the planWhich tasks are not yet assigned to any resource?
The AI answers from the planWho on the team can also take on the task {TASK}? Take into account: {CONTEXT}.
Add your own knowledgeRisk
3 prompts to copy
Which tasks have no buffer and would push back the project end if they slip?
The AI answers from the planWhich tasks depend on external parties according to the plan?
The AI answers from the planWhat options do I have if the building permit is delayed? Current status: {PERMIT_STATUS}.
Add your own knowledgeCosts
3 prompts to copy
What are the planned total costs, and how are they distributed across the phases?
The AI answers from the planWhere do the actual costs deviate most from the planned costs?
The AI answers from the planI have to cut the budget by {PERCENT}. Which tasks are candidates for this? Please take into account: {CONTEXT}.
Add your own knowledgeReports
2 prompts to copy
Create a status report for management: half a page at most, with a traffic light for dates, costs, and resources.
The AI answers from the planWrite a short, factual email to {RECIPIENT} about the current project status. What matters most to the recipient: {CONTEXT}.
Add your own knowledgePrompts with a dashed marker need information that is not in the plan. Replace the marked parts with your own knowledge. The more context you give, the better founded the assessment. You still make the decision.
Before you pass on a number, check it briefly in the plan. To try it out, use the sample project “Single-Family Home” from the Merlin Project templates.
How to connect your AI assistant to Merlin Project is explained in the article Which AI client is right for you? The full reference is in the MCP documentation.
If you have any questions about this blog article or would like to discuss it, we look forward to your contribution in our forum.
Frequently asked questions
Which questions does the AI answer directly from my project plan?
Everything that sits in the plan as a value: status and delay, dependencies and critical path, dates, durations, and assignments, as well as notes and attachments. From these, the AI also creates summaries and reports.
Why does the AI not give me a clear answer to some questions?
Because the necessary information is not in the plan, such as verbal agreements, the status at an authority, or priorities set by management. The AI says so openly, weighs things up based on the available data, and offers options. It does not make a binding decision.
What changes when I give the AI additional context?
The AI factors your information into its assessment and orders the options accordingly. In the video, one sentence on the status of the building permit is enough for the suggestions to fit the actual situation much better.
Where do I write the additional context?
The easiest way is directly in the prompt, for example: “According to the building authority, the building permit is being processed and will probably arrive next week.” If you need the information more often, enter it as a note on the matching task in Merlin Project. The AI then reads it on its own next time.
Does the AI then decide for me?
No. Even with context, the decision stays with you. The AI delivers a better-founded assessment, but it never knows every connection in your project.
What do I do if the AI gives a clear recommendation anyway?
Check the reasoning. For questions that need knowledge outside the plan, the AI then usually relies only on dates and dependencies from the plan. What it does not know does not go into the recommendation.
Can the AI adjust my plan when it suggests an option?
No. The Merlin Project MCP server only reads. The AI can tell you what happens if a milestone moves. Moving it is up to you.
Can I rely on the numbers in the answers?
Mostly yes, but check important values briefly in the plan before you pass them on. AI assistants can misread values or assign them to the wrong item.
How can I try the two questions from the video myself?
In Merlin Project, open the sample project “Single-Family Home” from the templates, share it via “AI Tools”, and ask the two questions through an AI assistant connected to the MCP server. Ask the second question once without and once with the sentence about the building authority, and compare the answers.
