Good morning Engineers,
OpenAI’s new Data agent brings company data and documents into a conversational workflow for analysis and dashboards. Our take at Singularity: the useful starting point is a tightly defined project question, supported by current records and a clear path for an engineer to check the answer.
Today: AI for project data, better meeting follow-through, and a prompt for preparing your next engineering review.
AI NEWS
1. Project data gets a new interface
On September 10, OpenAI introduced its Data agent in ChatGPT Work, which can connect approved company data and documents for analysis. For oil and gas project teams, a useful pilot could be identifying overdue vendor reviews from a controlled document register. Define what counts as overdue and check the returned records before using the result in a client update.
2. Meeting intelligence moves closer to action
Superhuman announced its acquisition of Fathom on September 14 and plans to bring meeting intelligence into its broader suite. Our engineering takeaway is to test how well meeting notes preserve decisions, owners, and due dates before feeding a project tracker. Treat the planned integration as a direction of travel, while evaluating Fathom’s current capabilities on their own merits.
3. More capable agents need defined access
Anthropic’s September threat report describes misuse in which AI took a larger role in coordinating cyber operations. For engineering firms connecting agents to project files, this reinforces the value of narrowly scoped permissions. Start a pilot with read access to an approved folder and require a reviewer before changes reach issued records.
Image: Fathom
TOOL SPOTLIGHT | FATHOM
Keep the decision behind the action item
Fathom turns meetings into searchable transcripts, summaries, and action items. For engineering coordination calls, use that record to prepare follow-ups and recover the context behind a decision. Check equipment tags, technical values, and assigned owners before circulating the minutes.
With Fathom, you can:
Capture a transcript and summary while you focus on the discussion.
Search past conversations for decisions and unresolved questions.
Use recorded action items to prepare a follow-up list for review.
FROM THE FRONTIER
The Data agent meets the document register
What changed: OpenAI’s Data agent connects approved data sources and can bring Drive or SharePoint documents into analysis. It can produce dashboards and support follow-up questions. Access follows the connected account’s permissions.
Why engineers should care: Consider a weekly vendor-document review. A defined query could identify late submissions, while approved correspondence provides context for the delay. The reviewer still needs the document number, revision, status date, and source behind each finding.
The market signal: Our interpretation is that organizing project information is becoming part of AI readiness. Consistent status definitions make an analysis easier to check. A folder containing several revisions still needs a rule for selecting the governing record.
A useful first test: Use a completed project with known answers. Compare an AI-prepared overdue list with the checked register, recording missed items and false flags. Measure preparation time plus correction time before deciding whether to repeat the workflow.
Image: Van Tay Media / Unsplash
LEARNING HOW TO MARKET
Show clients a workflow they can evaluate
Choose one recurring client problem, such as slow vendor-document turnaround, and describe the consequence in plain language.
Build a small demonstration using synthetic inputs or material cleared for external use, showing the starting point and reviewed output.
Explain the workflow through three visible stages: input preparation, AI assistance, and engineering review.
Report a measured result only when you have a documented baseline, comparable scope, and the time spent correcting the AI output.
Invite readers to talk with Singularity about testing the same workflow on a defined part of their project.
AI: Review complete.
Engineer: Which revision?
AI: The one called FINAL_final_v7_USE_THIS_ONE.pdf.
IN THE KNOW
✳ 7 workflow ideas: Scan Gamma’s seven-agent guide for candidate tasks, then test the relevant tool against your own acceptance criteria.
😄 From the archives: Anthropic’s Project Vend experiment is an entertaining reminder that running an office shop can challenge an AI agent.
▶️ Watch a workflow: Anthropic’s Social Content Engine demo offers a starting point for thinking about repeatable content preparation.
✍ Learn from Andrew Ng: His Agentic AI course covers reflection, tool use, and evaluation for people ready to build and test workflows.
☕ Reddit reality check: One electronics-workflow post claims extensive automation, but a personal demonstration is not evidence that a system can meet your project’s acceptance criteria.
PRODUCTIVITY SOFTWARE
5 New and Trending Marketing Tools
ChatGPT Data agent: Explore campaign or enquiry data through connected sources and dashboards to prepare a checked business-development update. Official release.
Claude for Small Business: Its newly expanded workflows include proposal and content preparation, offering technical firms a starting point for repeatable outreach. Official overview.
Fathom: Search meeting transcripts and prepare follow-ups after discovery calls or project discussions. Official product.
Krea Agent: Its creative-production beta can develop image and video assets from a brief for an approved campaign. Official guide.
Superhuman Docs: Combine a content brief, publishing tracker, and team feedback in a shared document with tables and AI assistance. Official product.
SPEEDING ENGINEERING WORKFLOWS WITH AI
Prepare a review list with evidence
Act as an engineering review coordinator. I will provide a document register, meeting notes, and relevant correspondence for [project name], with a reporting cut-off of [date]. Use only these inputs.
Prepare a review table with: document or equipment tag, document number and revision, issue requiring attention, supporting filename and page or row, stated owner, stated due date, and next action. Apply the governing-revision rule [insert rule]; if it is missing or revisions conflict, flag the item for clarification.
Keep confirmed decisions separate from proposals and open questions. Preserve units and technical values exactly as written. Write “not stated” for missing owners or dates. Do not invent approvals, engineering conclusions, or completion status. Finish with the three questions most likely to unblock the review. Return a draft for checking; do not modify the source records.
Result preview: You get a review list with traceable sources and visible gaps, ready for the project engineer to check. The next meeting can focus on unresolved decisions and missing inputs.
PUT IT TO WORK
Choose one workflow to improve
Which recurring review or handoff would you like to make easier for your team?
Explore Singularity’s AI engineering and automation capabilities to see how we approach practical workflow improvements.
Contact Singularity to discuss a defined opportunity in your project delivery process.
How useful was this issue?
See you next issue, with another practical way to improve engineering work.
Singularity Engineering Canada