TensorPM, Claude Code and Codex: Putting the project plan first

The project plan looks complete. Tasks have owners, dates are set and handover is scheduled for Friday. But what does “ready for use” actually mean? Who accepts the result? Has time been allowed for that? Will the planned work deliver the agreed outcome?

Even a detailed plan can leave those questions unanswered. A team can be busy while a crucial prerequisite is missing or people have different ideas of what success means. If those gaps only become clear just before handover, a long list of completed tasks offers little comfort.

In TensorPM, guiding the project is the central task. You and the AI examine whether the goals are clear, the plan has gaps or contradictions, and the work is moving forward. Where do we stand today? What have we actually achieved? What puts our goal at risk? A transparent project context provides the foundation for answering those questions and deciding what to do next. (Suggestions & Order)

TensorPM brings three parts of project work together: Understand, Steer, Execute.

• Understand: Clarify goals, requirements, decisions and constraints, and capture them in a shared project context.
• Steer: Check the plan for gaps and contradictions, assess status and progress, and decide what needs to happen next.
• Execute: Carry out work, research and analyse information, work on tasks and create documents.

These activities build on each other. The context supports analysis. The analysis helps you decide what matters now. Execution draws on those facts and decisions. New findings can be reviewed through the Distiller and used to update the project context. (How TensorPM works)

The project context brings together what matters to the work: goals, success criteria, requirements, dates, budget, risks, decisions and tasks. TensorPM gives each of these areas a place from the outset. That lets you review the plan as a whole and identify questions that still need an answer. (Project structure)

You start using that structure when you create the project. Describe what you want to achieve or bring in existing documents. TensorPM can prepare a first draft for you to check and develop. You work through the essentials: What are we trying to achieve? How will we know we have succeeded? What must we deliver? Which conditions need to be met? (Create a project)

If the brief simply says “ready for use by Friday”, someone needs to establish what that means. Does it include a passed inspection, trained staff and client approval? Once that is clear, you can assess whether the plan includes the work required.

TensorPM helps you spot gaps and contradictions early. Its project analysis can use the information recorded in the project to examine questions such as:

• Is the goal specific enough, and are the success criteria clear?
• Is important information missing or inconsistent?
• Are the planned tasks sufficient to achieve the goal?
• Which tasks are blocked, and which decisions no longer support the goal?

The findings help you ask focused questions and improve the plan. Training might be required, yet no task has been created for it. A necessary approval might be scheduled after handover. These gaps affect whether the project can succeed, which is why checking for them is central to TensorPM's work. (Project analysis)

Current status and progress matter just as much. Progress tells you how much of the intended outcome has been achieved and what remains. Status describes the situation today: Is the project on track? Where is it blocked? What threatens the dates or goals? TensorPM produces AI assessments based on the recorded project information. Your own status assessment remains separate, so you can compare your judgement with the AI's. (Status and progress)

A long list of completed tasks can look encouraging. Yet the project may still be unready for handover if a crucial inspection or approval is missing. Progress and status need to be considered alongside the goals, success criteria and outstanding prerequisites.

For the next project meeting, you need answers to practical questions: What have we achieved since the last update? What remains before acceptance? Where might delays occur? Which decision is needed now? The maintained project context gives you and the AI the information to work through those questions.

The AI can miss something. It also cannot account for information that exists only in your head. The value comes from working together: you contribute experience and missing facts; the AI helps you examine the plan, check connections and identify open questions. Run the analysis again after significant changes.

You and the AI develop the same project information together. You can clarify a goal, add a risk or correct an assumption. The AI uses that information in its work. You can see what has been recorded and discuss the basis for a recommendation. Keeping the goals clear becomes a shared task throughout the project.

This is TensorPM's focus in comparison with coding agents such as Claude Code and Codex. Those tools can handle demanding tasks, plan their approach and check the results. TensorPM directs its structure and analysis towards the whole project: Is the plan complete enough? Does it make sense? How far have we come, and are we still on track? That purpose shapes the application's everyday workflow. (Claude Code, Codex)

Ongoing management of an entire project is not the product focus of Claude Code or Codex. Their strength is carrying out work assignments. They do not come with a ready-made project management process that combines a fixed structure for project information, regular reviews of the plan, assessments of status and progress, and human approval of context updates. This is an assessment of their documented capabilities and their focus on getting work done. (Claude Code, Codex)

For that use, the structure and process need to be set up separately. Which information belongs in the project? How are open questions recorded? When is the plan checked again for gaps and contradictions? How do new commitments become part of the agreed project record? And how does that information lead to an assessment of status, progress and next steps that the team can understand?

The team has to build and maintain that process with its own instructions, workflows and any additional tools it requires. It must establish how project knowledge is created, who reviews changes and how the record stays current. A good one-off review of a plan does not ensure that later changes will be checked consistently and considered in the next decision.

Claude Cowork and ChatGPT Work explicitly address knowledge work beyond programming. They take on delegated assignments, conduct research, analyse information and produce deliverables such as reports, spreadsheets and presentations. They can also run recurring workflows and organise related work. Their focus is on the assignment and its outcome. (Claude Cowork, ChatGPT Work)

TensorPM also supports research, analysis and the creation of deliverables directly within the project. Its project agent can conduct research, analyse information, draft emails and create documents, spreadsheets or presentations. Document creation tools are built in. You can delegate concrete work using the project context you have already developed. (Project work and results, Built-in document tools)

Alongside the supplied files, the shared project context holds clearly organised goals, success criteria, decisions, risks, tasks and the current situation. A report can draw on what has been agreed, what has been achieved and which decisions remain open. A presentation for the next project meeting can use the same information. The outputs stay connected to the work of managing the project.

A status report or an updated presentation can be valuable to a project. Ongoing management also needs a shared project record that lets the team review how an approved change affects the plan, risks, open tasks and the next assessment of status and progress. Organising assignments, results and recurring workflows together does not, on its own, establish an end-to-end project management process. How changes affect the plan and its assessment still needs to be defined.

In TensorPM, this structure and process are part of the core product. They connect initial goal-setting with day-to-day project management. You work with the AI on the plan, review what has been achieved and decide which proposed changes to accept. Reviewing, completing and improving an initially incomplete plan becomes part of everyday project work. (How TensorPM works)

Claude Code and Codex can also retain information from earlier interactions that seems useful for later work. The AI selects what to keep. By itself, that memory does not provide a fixed structure for checking a project plan for missing goals, requirements or tasks. (Claude Code memory, Codex memories)

That selection can matter. A supplier writes: “Friday may be possible, provided the inspection is finished in time.” If only “delivery possibly Friday” is retained, an important condition is absent from the next decision. The project manager needs the relevant facts and unresolved questions to remain part of the shared project information.

The Distiller brings new information into that context through a defined review process. It reads incoming emails, documents and notes in the context of the project, then proposes what should change. You can amend, accept or reject those proposals. Changes enter the project record when you approve them. (Files, intake & trail)

You check that the proposal includes important facts and conditions and distinguishes confirmed information from uncertainty. This human review, often called “human in the loop”, is built into the process. If the AI overlooks an important point, you can add it before approving the update.

For the supplier's message, the first step might be a follow-up question. “Friday may be possible” is not a commitment. You can reject a premature date change and ask the supplier when they can give a firm answer.

If Friday is later confirmed, you can approve the update. Then you and the AI review the plan again: Is there still time for setup and acceptance? Which tasks need to move? Who needs to decide what happens if the original handover date is no longer feasible? The new information becomes part of managing the project.

TensorPM connects building the plan with reviewing and updating it throughout the work. A risk, a changed requirement or a new decision can prompt you to reconsider the next steps. The context brings together the information you need to make that judgement.

When the AI takes on tasks, its access to your computer matters too. Coding agents can launch programs and edit files, with safeguards that limit those capabilities according to configuration and permissions. TensorPM's project agent uses tools intended for project work; running arbitrary commands on your computer is not part of its standard capabilities. Calculations and file creation take place in protected areas with defined access rights. (Codex security, Claude Code sandboxing, TensorPM tools)

For suitable software tasks, you can also use Codex or Claude Code through TensorPM. The task stays connected to the project while the external agent handles the technical work. Those connections operate under their own access permissions. (Agent integrations)

TensorPM brings these activities together: clarifying goals, checking the plan for gaps and contradictions, assessing status and progress, and incorporating new information with human review. Together, they inform the next steps. People and AI work with a visible, structured project context throughout.

Understand, Steer, Execute describes the full workflow. TensorPM helps you understand the project, guide it towards its goals and carry out the work ahead. The same shared context supports each step.

A clear plan and an up-to-date picture of the project show what you are trying to achieve, how far you have come and what needs to happen next. They give you a foundation for moving the project forward with purpose.

The introduction to TensorPM walks you through the process, from setting up your project context to using it in your daily work.