In 2001 the Agile Manifesto asked teams to deliver working software frequently, from a couple of weeks to a couple of months, with a preference for the shorter timescale. Twenty-five years later, working with an AI assistant means that shorter timescale is measured in hours. You write what you need, the AI returns a complete first draft, you look at it, correct it and ask again. That is agile, even if nobody calls it that.
But speed alone is not agile. Without a clear goal, a definition of done and someone who verifies, AI produces a lot and delivers little. This module closes the seminar with that idea: AI does not replace project management; it forces it back to the essentials. It picks up the four previous modules and carries them into the way products, documents and analyses are built today.
What the manifesto asked for and AI made normal
Module 4 explained the manifesto’s four values. Read with AI, their weight shifts:
| The manifesto values… | …over | What changes with AI |
|---|---|---|
| Individuals and interactions | Processes and tools | The conversation with the AI is the process: knowing how to ask and review is worth more than mastering the tool |
| Working software | Comprehensive documentation | A prototype you can open costs about the same as describing it; you show instead of explaining |
| Customer collaboration | Contract negotiation | The customer sees a version the same day and reacts to something real, not to a document |
| Responding to change | Following a plan | Changing course is cheap; the plan is revisited at every delivery |
The sentence almost nobody reads in the manifesto still holds: the items on the right also have value. With AI that warning matters more, because producing documents, code and slides is no longer expensive. The risk is no longer doing too little; it is doing too much.
The six principles
The first three come from agile as it always was. The other three are the ones AI makes mandatory.
- Value leads, not the task. Every cycle starts with a question: what problem does this solve and how will I know it did. If there is no answer, nothing gets built. It is the business case from module 1, in miniature.
- Small, frequent deliveries. A page, a filter, a spreadsheet tab. Something the user can open today and comment on tomorrow.
- Change is raw material. If changing is cheap, the plan is adjusted at every delivery, not at the end of the project.
- Verify before delivering. AI writes fast and is wrong with confidence. Every figure is checked against the source, every piece of code is run and every page is opened before anyone says it is ready.
- Human judgment is not delegated. AI proposes and executes; the person decides what is worth doing, which risk is accepted and what is not published. Ethical and legal rules, such as Colombia’s personal data law (Law 1581 of 2012), are written down from day one.
- Memory belongs to the team, not to the conversation. Decisions, mistakes and verified figures are written down so the next cycle does not relearn them. It is the Scrum retrospective turned into a file.
Principles 4 and 5 give a practical rule for using AI on any project: AI transforms what exists; it does not invent what does not exist. It can write the executive summary from the business case, the charter from the approved case, or the status report from the risks and the progress. It cannot estimate benefits, costs, durations or discount rates, nor decide whether a project is green or red.
Principle 4 is the one most often neglected. A well-written answer looks correct, and that is the trap: form no longer signals quality. The only defense is the same one Scrum uses, a Definition of Done that is not negotiable.
The five-step cycle
A cycle lasts from a few hours to a week and always has the same steps. When the fifth closes, you go back to the first with what you learned.
- Goal. One sentence with the expected result and how it is measured. It is the Sprint Goal, in miniature. Example: a company’s profile shows its public contact details by year.
- Rules. What must not be done: use personal data, publish figures without a source, promise what will not be delivered. The AI works inside that frame.
- Build with AI. The AI produces the complete first draft: code, text, spreadsheet or analysis. The person reviews the approach, not every line.
- Verify. You open what was built, check figures against the source and test it with a real user. Whatever fails goes back to step 3. Whatever is almost ready is not delivered.
- Deliver and record. It is published, you write down what was decided and which mistake not to repeat, and you pick the next goal based on what the user said.
Compared with the Sprint
| Scrum Sprint | AI cycle | |
|---|---|---|
| Length | One month or less | Hours to a week |
| Who builds | The Developers | The AI, directed by a person |
| Who verifies | The team, with the Definition of Done | The person, with the same rule |
| Where learning is kept | The retrospective | A memory file that the next cycle reads |
What does not change is what matters most: without verification there is no delivery.
Before the first cycle: why the project exists
The five-step cycle answers how to build fast. It does not answer whether the project should exist. That question comes first: a strategic objective, an initiative, a business case and an approval with a budget. Without them, AI only speeds up something nobody decided to do. That is why step 1, the goal and its measure, comes from the approved business case, and step 5, the memory, is kept next to that same case, so that at the end you can answer whether it was worth it.
What does not change
AI speeds up execution. It does not replace management.
- The PMBOK is still the map. The five process groups are still there: module 2 explained Initiating and module 3 completed the tour. The charter, the stakeholders, the scope and the WBS are still needed; what changes is that AI helps draft them in hours.
- The life cycle is still a choice. As module 2 showed, whatever is regulatory or contractual is planned in full; whatever the user discovers by using it is worked in short cycles.
- Lean still measures. With AI, the typical waste is building what nobody asked for, and the typical variation is a figure that changes from one version to the next.
- Governance is not negotiable. Who approves, what is published and what stays confidential is decided by a person. A compliance error costs more than any automation saves.
The full map: the PMBOK and agile, piece by piece
The practical question is not whether to use AI, but in which part of the project it helps and in which it does not. To answer it, we walk through everything the seminar covered: the organization’s strategy, how the project is born, the life cycle, the five process groups, the ten knowledge areas, the performance domains and the agile frameworks. For each piece, one column says what AI can do and another what a person still decides.
The organization’s strategy (module 1)
A project that does not hang from the strategy is an expense. The cascade from module 1 is unchanged; what changes is how quickly it can be reviewed.
| Level | What AI does | What the person decides |
|---|---|---|
| Strategy | Summarizes the market, the competition and internal figures to prepare the discussion | The company’s direction and what it will not do |
| Portfolios | Compares candidate projects by benefit, cost and risk, using the assumptions it is given | What enters the portfolio and what is cancelled |
| Programs | Shows dependencies between projects and benefits only achieved together | How they are coordinated and who answers for the benefit |
| Projects | Links each project to the objective and the indicator it should move | Whether the project still deserves the money |
| OKRs and KPIs | Calculates progress on key results and flags when a KPI goes out of range | The quarter’s objectives and what to do when an indicator drops |
With AI it is tempting to launch many projects because each one looks cheap. Strategy exists precisely for this: to say no. The time of the people who review and decide is still scarce, even if building no longer is. The link between strategy and execution is covered in OKRs and KPIs: one leads, the other reports.
How a project is born (module 1)
| Station | What AI does | What the person decides |
|---|---|---|
| Process, initiative or project? | Helps sort a list of pending items by start, end and deliverable | What is formalized as a project and what stays as operations |
| Link to strategy | Relates the idea to the company’s objectives and OKRs | Whether it truly hangs from the strategy or is a pet idea |
| 1. Problem identification | Organizes complaints, data and emails to show where the pain is | Which problem is worth tackling |
| 2. Proposed solution | Generates several alternatives and describes the ideal future process | Which alternative moves forward |
| 3. Feasibility assessment | Prepares the questions for the technical, legal and financial experts | The experts’ answers do not come from AI |
| 4. Business case | Drafts the executive summary from the other sections and calculates NPV, IRR, ROI and payback over six years with the assumptions it is given | The assumptions: benefits, costs, durations and discount rate, defended before the committee |
| 5. Approval | Prepares the presentation for the committee | Who presents, what is negotiated and whether it is approved |
With AI, a business case that used to take weeks comes together in hours. That does not make it better; it makes it easier to inflate. The assumptions are still the part someone defends with their own name.
The life cycle (module 2)
| Approach | With AI |
|---|---|
| Predictive | AI helps write complete requirements and the WBS, but scope is still fixed by a regulation or a contract |
| Iterative | Prototyping is so cheap that iterations to understand the product are counted in hours |
| Incremental | Each usable delivery arrives sooner, as long as it passes verification |
| Adaptive | The natural approach for working with AI: goal, draft, review and adjust |
| Hybrid | The most common case: the regulatory part is planned in full and the rest is worked in short AI cycles |
Module 2’s two questions still decide: whether the requirements are defined and will hold, and whether the business receives something usable before the end. AI does not change the answers; it changes how cheap it is to test before deciding.
The four gates
Between the idea and the closing there are four moments when someone with authority signs. With AI you reach each one faster, but none can be skipped.
| Gate | What happens | With AI |
|---|---|---|
| G1 · Business case approved | The committee assigns a budget and the initiative becomes a project; the case is frozen as a baseline to measure against later | AI assembles the committee package; the decision belongs to the committee |
| G2 · Charter signed | The sponsor formally authorizes the project | AI pre-fills the charter from the approved case; the signature belongs to the sponsor |
| G3 · Baseline approved | Scope, schedule and cost are fixed to control changes | AI proposes the WBS and the schedule; the commitment belongs to the project manager |
| G4 · Formal acceptance | The customer receives the deliverables and the project closes | AI prepares the closing document; acceptance belongs to the customer |
In an agile project G3 is lighter: the baseline is the Product Goal and the ordered backlog, not a full schedule.
The five process groups (modules 2 and 3)
Module 2 explained the Initiating group, together with the life cycle, and module 3 completed the tour. With AI they are still the same five:
| Group | What AI does | What the person decides |
|---|---|---|
| Initiating | Drafts the charter and the business case; lists possible stakeholders from documents | Whether the project goes ahead, who the sponsor is and what value it must produce |
| Planning | Proposes the WBS, an initial schedule and a risk register; turns the scope into user stories | The estimates that are committed, the budget and what is in or out |
| Executing | Builds drafts: code, documents, analyses, presentations | What is accepted, what is redone and what is published |
| Monitoring and controlling | Summarizes progress, compares against the baseline, spots figures that do not add up | Which deviation is acceptable and which change is approved |
| Closing | Assembles the final report and a draft of lessons learned | The customer’s formal acceptance and which lesson becomes a rule |
In an agile project, as module 4 showed, planning, executing and controlling repeat inside every Sprint. With AI they repeat inside every cycle of hours.
The ten knowledge areas (PMBOK, 6th edition)
| Area | What AI does | What the person decides |
|---|---|---|
| Integration | Keeps the plan, the change log and the project memory up to date | How scope, time and cost are balanced when they clash |
| Scope | Breaks deliverables into a WBS and stories; spots what was left ambiguous | Where the scope ends and what stays out |
| Schedule | Sequences activities, computes the critical path, proposes scenarios | The date promised to the customer |
| Cost | Builds the base budget and compares actual spend against plan | The approved budget and the reserves |
| Quality | Reviews against a checklist, tests code, checks figures against the source | The Definition of Done and the level of quality required |
| Resources | Calculates workload and availability | Who does what, and how the team is looked after |
| Communications | Drafts reports, minutes and summaries for each stakeholder | What is said, to whom and when, especially when the news is bad |
| Risk | Lists risks from similar projects and from the documents | Which risk is accepted, avoided or transferred |
| Procurement | Compares proposals and summarizes contracts | Who is hired and what is signed |
| Stakeholders | Maps stakeholders from org charts and emails | Who carries weight, who opposes and why; that is in no document |
The pattern repeats in all ten: AI prepares, the person commits. Estimates, budget, required quality and stakeholder relationships are commitments, and a commitment is not delegated.
The eight performance domains (PMBOK, 7th edition)
The seventh edition organized the standard by domains instead of processes. Here is how they look with AI:
| Domain | What changes with AI |
|---|---|
| Stakeholders | There is more to show and sooner; the risk is overwhelming the customer with versions |
| Team | AI becomes one more member who executes; leadership is still human |
| Development approach and life cycle | Prototyping is so cheap that almost every project has an adaptive part |
| Planning | You plan more often and with better information, just as in Scrum |
| Project work | The work shifts from producing to directing and reviewing what is produced |
| Delivery | Smaller, more frequent deliveries |
| Measurement | Measuring becomes cheap; the hard part is choosing what to measure |
| Uncertainty | A new risk appears: errors that look correct |
The eighth edition, from November 2025, again presents processes organized in the same five groups, applicable to predictive, adaptive and hybrid projects. The tables above still apply.
The agile frameworks
| Framework | Piece | With AI |
|---|---|---|
| Scrum | Product Owner | Still orders the backlog; AI helps write and split stories |
| Scrum Master | Makes sure the cycle is respected; with AI, above all that verification is not skipped | |
| Developers | Direct and review what AI builds; they answer for quality | |
| Sprint Planning | The goal is written the same way; the forecast changes because building is faster | |
| Daily Scrum | Still for adjusting the day’s plan, not for reporting to the boss | |
| Sprint Review | More is shown, and sooner; the customer reacts to something that works | |
| Retrospective | What was learned is saved in a memory file that the next cycle reads | |
| Backlog, Increment and Definition of Done | The Definition of Done becomes the most important piece | |
| Kanban | Limiting work in progress | More necessary than ever: AI lets you open ten fronts at once |
| The four measures | Work in progress, throughput, work item age and cycle time calculate themselves | |
| XP | Pair programming | The pair is now a person and an AI; one reviews what the other writes |
| Tests before code | The cheapest way to verify what AI programs | |
| Small releases and simple design | AI tends to overbuild; simple design has to be asked for | |
| Lean | Eliminating waste | The new waste is building what nobody asked for because it was easy |
| Reducing waiting | The waiting is no longer in building, but in approving and verifying |
The tables add up to one reading: AI increasingly takes care of the how; the what, the why and when to stop still belong to the person. That is exactly what agile always asked for.
The closing: when the project becomes a process
The seminar began by separating project from process: a project has a start and an end; a process repeats and does not end. And that is where it ends too. Every successful project dies by becoming a process: the new software moves to support, the dashboard becomes Monday’s report, the platform becomes daily operations.
It is the same move we saw with OKRs and KPIs: the OKR pushes something that does not exist yet and, once achieved, disappears; what remains watching it is a KPI. The project is the OKR; the process it gives birth to is the KPI.
| At closing | What AI does | What the person decides |
|---|---|---|
| Handover to operations | Drafts the operations manual, FAQs and support guide from the project memory | Who owns the process from the next day on |
| Process indicators | Proposes KPIs with formula, baseline and threshold, and builds the dashboard that watches them | Which threshold triggers an alert and who responds |
| Lessons learned | Organizes the project memory: decisions, mistakes and what must not be repeated | Which lesson becomes a rule for the next project |
| Formal acceptance | Prepares the closing document and the benefits summary against the business case | The signature: whether the project delivered what it promised |
| Business case follow-up | Compares realized benefits against those the case promised, year by year | Whether the project was worth it and what to learn for the next case |
| Continuous improvement | Reviews the process with data and proposes adjustments | When an adjustment is a process improvement and when it is a new project |
Closing the project is not the end of the accounting. A business case usually promises benefits over several years, and many arrive after closing. That is why the case approved at G1 is kept as a baseline and compared against reality once the project is already a process. It is the only way to answer the question the committee asked at the start: was it worth it.
With AI a risk specific to closing appears: something that works but nobody knows how it works. If the project was built in fast AI cycles and the memory was never written down, the process that operations inherits is a black box. That is why the sixth principle weighs more at the end than at the start: the project memory becomes the process manual.
And the circle closes. When one of that process’s KPIs goes out of range and continuous improvement is no longer enough, an initiative is born, which may become a new project: back to module 1.
Two cases built this way
This method did not come from a book. It came from building real products in parallel during October 2026, with 3 to 5 hours a day for both projects combined, always with the same pattern: a minimal version in front of a real user early, and adjustments based on what that user did with it. Two more prototypes, each with its own website, came out of the same period. Both cases in this section are already in use.
WiseData: public company data
wisedataco.com puts the public figures of Colombian companies within anyone’s reach: 32,817 entities searchable in three directories (companies, the cooperative sector and the financial sector), sector dashboards and downloadable tools such as the OKR and KPI System.
| Cycle step | How it looked at WiseData |
|---|---|
| Goal | One analysis or one feature per week, with a sentence stating which question it answers |
| Rules | Never personal data of individuals, every figure with its source and cut-off date, no investment advice |
| Build with AI | Processing Supersociedades financial statements, building the directories and writing the articles |
| Verify | Checking every figure against the official source; that is how we found companies reporting in pesos instead of thousands, and others reporting by half-year |
| Deliver and record | Each correction was written down as a rule, so the next analysis would not repeat the error |
Salon Pass: a platform for salons
salon-pass.com is a platform for hair salons, barbershops and spas, with its app at app.salon-pass.com. It was built with the salon open: every version was tested in the day’s operation and adjusted with what the people using it asked for.
| Cycle step | How it looked at Salon Pass |
|---|---|
| Goal | Solve one concrete need of the salon per delivery, not the whole platform at once |
| Rules | The salon’s customer data is protected; payments go through an established payment gateway |
| Build with AI | The app, the booking screens and online payments |
| Verify | Test it in the salon with real appointments and payments before calling it done |
| Deliver and record | What the salon asked for defined the next delivery |
How long it took
Salon Pass is the best-measured case. Since 1 October it has taken between 80 and 110 hours of the author’s time, and in that time it gained 33 database changes, about 20 server functions, the website, the app, online payments and the migration of real data from the previous system with a rollback plan. As a reference, and this is our own estimate, not a measurement, a team of two or three developers working the traditional way would have needed six to nine months, on the order of 2,000 to 3,500 hours, to reach the same point.
The difference is not explained by AI’s speed alone. It is explained by the cycle: every few hours there was something tested in the salon, and nothing was built without a written goal or called done without being verified.
Neither of the two started with a twelve-month plan. Both started with a concrete need and a first delivery within hours of work. And in both, the costliest mistakes were in verification, not in building: a figure the source reported in a different unit, a piece of data that should not have been shown, a link that did not open.
The most common mistakes
- Confusing speed with progress. Ten versions in a day are useless if none was tested with a user.
- Not writing the goal. Without a one-sentence goal, the AI fills the gap with whatever it thinks is useful, and it is almost never what was needed.
- Skipping verification because the text looks good. It is the most expensive mistake and the most frequent.
- Not keeping the memory. Every new conversation starts from zero and repeats the mistakes of the last one.
- Delegating judgment. Asking the AI whether something is right is not verifying.
The complete seminar
| Module | Question it answers |
|---|---|
| 1 · How a project is born | Why is this project being done and who approves it? |
| 2 · How to choose a project life cycle | Is everything planned up front or discovered along the way? |
| 3 · The five process groups and where AI fits in | What has to happen for the project to end well? |
| 4 · Agile and Scrum | How do you work when the problem cannot be fully planned? |
| 5 · AI is the new agile | How do you work when AI executes and the person decides, from the project’s birth until it becomes a process? |
The module challenge
Pick a real task from your work that takes you more than a day today and plan one cycle:
- The goal in one sentence, with how you will know it was met.
- Three rules: what the AI must not do or use.
- What you will ask the AI for as a first draft.
- Your verification checklist: what you will check, open or test before calling it done.
- What you will write down at the end so the next cycle goes better.
If you cannot write the goal in one sentence, that is the finding, just as in module 4: the problem is not the tool, it is that what you want is not yet clear.
A note on how this was written
This article was written with the method it describes: goal, rules, a draft with an AI tool, verification and memory. The structure and the cases come from the author’s professional experience and from the two cases mentioned. The values and principles of the Agile Manifesto are presented as in module 4.
PMBOK is a registered mark of the Project Management Institute, Inc. This material is educational and is not sponsored or endorsed by PMI.
Sources
- Manifesto for Agile Software Development — Beck, K. et al. (2001)
- Principles behind the Agile Manifesto
- The Scrum Guide, November 2020 — Ken Schwaber and Jeff Sutherland
- PMBOK Guide, 6th edition (2017) and 7th edition (2021) — Project Management Institute
- The Kanban Guide, May 2025 version
- PMBOK Guide, 8th edition (November 2025) — Project Management Institute
- Law 1581 of 2012, personal data protection (Colombia)
The series on YouTube
This topic has its own playlist on the channel: the videos in order, to watch end to end or pick up where you left off.