
Supercharging Project Management with AI at DaySpring Limited Program
Supercharging Project Management with AI at DaySpring Limited Program
Today I gave a presentation on Supercharging Project Management with AI & MCP at the AI in Project Management event organised by DaySpring Limited — alongside my co-presenter Md. Mahmudul Hassan (SAFe® SPC, LPM, LSSMBB), both of us from AI4BT.
Here is the thing I wanted the room to leave with: project managers do not lack data. They lack a way to assemble it.
The status update is in the project tool. The risk is in a comment thread. The decision is in meeting minutes. The budget truth is in the ERP. Six systems, one PM, no connectors — so the PM becomes the integration layer, rebuilding the same picture by hand, every week, forever.
MCP is the plug. The AI is the reader.
I kept the technical part short, because the business idea fits in one image: MCP (Model Context Protocol) is USB-C for AI.
One standard plug between the AI and your systems. Not a custom integration per tool, not a new API project every time someone asks a question. MCP is an open standard under the Linux Foundation, and the major AI tools already speak it.
Once that plug exists, the questions change. Instead of "export the board to Excel and I'll build a summary", you ask:
- "Summarise all tasks and status for Project 2."
- "Which task has a deadline tomorrow, and who owns it?"
- "Which sub-tasks have blockers or dependencies?"
- "Who has the least workload right now?"
- "Which tasks are most likely to delay the project?"
The demo: Odoo + MCP + DeepSeek + goose
We ran these live, in both English and Bangla — because if a tool only works when you phrase the question in perfect English, your team will not use it.
The stack was deliberately unglamorous and cheap:
- Odoo as the system of record — projects, tasks, sub-tasks, owners, deadlines, dependencies
- MCP server as the plug between Odoo and the model
- DeepSeek as the reasoning model
- goose as the open-source agent running on the machine, doing the actual work
No screenshot-to-dashboard ceremony. Plain language in, structured answer out, every claim traceable to its source record.
That last part is the whole argument. Nothing is invented. The AI does not guess the status; it opens the record and reports it.
The ten challenges we walked through
- Getting a complete status report — data lives in boards, email and chat; the PM rebuilds it by hand
- Weekly and monthly reporting — same figures copied, same narrative rewritten
- Chasing missing updates — the PM becomes the reminder engine
- Explaining why something is late — the reason sits in comments nobody reads
- Tracking actions after meetings — decisions typed in by hand, or quietly lost
- Spotting risk early — overdue work found only by manual scanning
- Checking workload and availability — assignments, leave and calendars in different places
- Managing scope change — requests arrive informally, never priced against the plan
- Reconciling budget, time logs and progress — compared by hand, month after month
- Answering the same stakeholder questions — every stakeholder wants a different cut
Every one has the same shape: the information exists, the assembly is manual. That is the gap an MCP-connected agent closes.
The part executives actually care about
Two guardrails, stated plainly on stage:
The AI drafts. The human sends. Follow-ups, client reports, change-impact assessments — all land as proposals a PM authorises. Autonomy is a setting, not a default.
Permissions decide the answer. A client sponsor must never be answered with internal margin data. Access control is the control — not an afterthought.
And on cost: the capable open-weight models are good enough for this work. Choosing DeepSeek and an open-source agent over the most expensive option is not a corner cut — it is how the second project becomes cheaper than the first.
What I took away from the room
The best question of the day was not about the model: "If my team could ask our own system anything, what would we finally stop doing by hand?"
That is the right question. Not "which AI should we buy" — but "which manual assembly job are we still paying a senior person to do every week?"
Thank you to DaySpring Limited for organising, to Md. Mahmudul Hassan for sharing the stage, and to everyone who stayed back to ask questions in Bangla and English.
If your project status still lives in five systems and one person's memory, let's talk.
Sadiq Mohammad Alam
Metamorphosis Ltd. · AI4BT
sadiqalam.com
#ProjectManagement #AI #MCP #Odoo #EnterpriseAI #DeepSeek #AIAgents
Интересуетесь подобными мероприятиями?
Свяжитесь с нами, чтобы пригласить Садика М. Алама на ваше следующее мероприятие или обсудить потенциальное сотрудничество.
Свяжитесь с нами