Most people only meet GenAI in a chat window. You type something. It types something back. That is a good start. It is also a small part of what you can already do, often inside the subscription you already pay for. This article explains the six jobs, how to reach them, and where they fit in an R&D project.
Chat is only the start
Chat already does more than most people think. You ask a question. Before it answers, it can search a website, look inside a document, or run a calculation. One chat often uses several of these jobs. You do not see the switch.
Here is a simple example. Open NotebookLM, or a similar tool. Upload a few documents. Ask it to turn them into a podcast. You get a two-person conversation, with script and audio, from those documents. That is more than predicting the next word. The tool reads the source, structures it, writes the dialogue, and generates the audio. One line in the chat box starts all of that.
The chat window is the way in. The same window uses different jobs, depending on what you ask. GenAI has six jobs.
The six things generative AI can do
1. Finder — the colleague who remembers everything
You search your company files for something you know is there. You cannot find it, because the author used different words. Finder searches by meaning, not by the exact words. Ask “have we had this XXX problem before?” Finder brings up the right report, even when the file has a different name.
2. Organizer — the tidy-up crew
Every company has the same mess. Five people, or five systems, use five names for the same material. Organizer matches them to one common name, so you can use the data together. It can clean structured data. It can also put structure into documents. You need this job to make data Findable, Accessible, Interoperable and Reusable (FAIR).
3. Reasoner — the sharp colleague who pressure-tests your ideas
You want to try something new. Use Reasoner. It checks the idea against cost, safety limits, and rules. It says whether the idea fits. It also says why.
4. Author — your own personal assistant
Author drafts your email, report, summary, or SOP from your project data. You do not start from a blank page. You do not have to hunt for the template. You still review the draft and finish it. It helps when you are stuck starting a new document.
5. Checker — the second pair of eyes
A document is about to go to a reviewer or a regulator. Checker looks for what is missing or inconsistent. These are the gaps you miss when you proof-read your own work. Checker does not sign the document. It catches the basic gaps first, so the reviewer does not have to.
6. Orchestrator — the one who brings it all together
People forget this job. It is the most valuable one. Finder found the information. Reasoner weighed it. Author wrote it up. Checker confirmed it. Orchestrator runs those jobs in order, and in loops, until you have the result. It connects the steps. This is a role NotebookLM is playing when it manages the process of converting your document into a podcast.
Where to find each role in Microsoft 365 and what gets in the way
Most companies use Microsoft 365. Here is where each job sits in Copilot today, and why teams are often disappointed.
| Role | Where it lives in Microsoft 365 / Copilot | What usually causes disappointment |
|---|---|---|
| Finder | Copilot searches SharePoint or OneDrive by meaning. The search is generic, so it covers many document types. It also returns a limited number of results. | A messy SharePoint library is hard to search. Files are duplicated and named inconsistently. Copilot then misses the best documents, and the answer is weak. |
| Organizer | Copilot can structure your data if you give it a clear instruction. | To do this well at scale, you need stronger instructions and an automation pipeline. The latter is especially needed when you are fixing a whole dataset. |
| Reasoner | In chat, Copilot can weigh real constraints if you give them: specs, cost limits, supplier data. A general question is not enough. | Give Copilot your own document with the constraints. Stop it from using generic files in SharePoint or OneDrive. |
| Author | In Word, Outlook, and PowerPoint, Copilot drafts from content you attach. | On the default settings, the draft is generic. It does not follow your team's report format. Saved instructions or a template fix this. Teams rarely set them up. |
| Checker | In chat, Copilot can check an attached document against your team's checklist. A general “review this” is not enough. | Give Copilot your own checklist. Stop it from checking against generic or guessed rules. To check many documents, you need an automation pipeline, not a chat. |
| Orchestrator | Microsoft 365 Copilot cannot do this. An Orchestrator would update a record, notify someone, or start a process when a condition is met. Chat can describe the next step. Chat cannot take it. | The chat window does not have the access or the permission for this job. A person still has to run the steps. You need a different tool. |
Copilot natively is a Finder and Author. With clearer instructions, Copilot can also be an Organizer, a Reasoner, or a Checker. To run those three across a whole dataset you need an automation pipeline. Copilot is not an Orchestrator. Microsoft sells Copilot Studio for that. It is made for technology-savvy users though.
How a coding agent works
A coding agent uses these six jobs when it writes code. As Finder, it searches the codebase by meaning. Ask “where do we already do this?” and it finds the right function, even under another name. As Organizer, it keeps the names and structure the repository already uses. As Reasoner, it checks the approach against what must not break. When a run fails, it says why. As Author, it writes new code, or it suggests changes to code that is already there. As Checker, it looks for what is missing or inconsistent before you accept the change. As Orchestrator, it runs the test, the build, and the page. If the result is wrong, it goes back through Reasoner and Author until the change works.
Generative AI opportunities across the innovation funnel
A typical R&D project has six phases: ideation, feasibility and concept screening, bench-scale development, scale-up and pilot, validation and regulatory, and launch and commercialization. GenAI can help in all six. Early on, it finds things and matches patterns. In the middle, it drafts and keeps documents consistent. At the regulatory stage, it checks that nothing is missing. At launch, it helps with coordination. Here are four major opportunities for GenAI across the innovation funnel.
1. Turning market signals into product opportunities (Finder and Author)
Pain point: There is a huge amount of scientific content, market trends, consumer feedback, and competitor intelligence documents. Turning that into a clear set of actions is hard.
How GenAI helps: Finder gathers these signals. Author turns them into themes and avenues for new products. The team then starts from the whole picture.
2. Checking against regulatory requirements (Reasoner, Checker and Orchestrator)
Pain point: Teams find gaps against regulatory requirements late. The fixes are then costly.
How GenAI helps: Checker compares a draft with the requirements and lists what is missing. Reasoner explains why each item matters. The reviewer can then focus on judgment. Orchestrator can run this as an agent.
3. Reports and administrative documents (Finder and Author)
Pain point: Scientists lose time writing logs, stage-gate documents, and other reports in every phase.
How GenAI helps: Finder pulls the relevant data. Author drafts the document in your team's format. The scientist reviews it and finishes it.
4. Extracting data from diverse sources (Organizer)
Pain point: Lab data is often hard to pull together. It sits in lab experiment records, instrument exports, reports, and spreadsheets, each in a different format.
How GenAI helps: Organizer pulls the data out and puts it in one format. You can then quantitatively analyse it.
How far your team can go depends on three things. Which GenAI tool you use, and what it can do, and whether it is connected to your own data. And most important of all, how clean and well organized your underlying data is.
GenAI is a multifaceted technology that can do much more than just chat. It can help automate many tedious R&D tasks, and make room for more value added ones.
Frequently asked questions
Do I need a special tool to get started with generative AI in R&D?
No. A chat with your own documents attached is enough to draft, reason through a problem, or check one document against a checklist. Searching your company files needs a Finder connected to them, and working with live systems such as LIMS, ERP, or PLM needs an Orchestrator with that connection.
Why are results from Copilot or another enterprise generative AI tool disappointing in R&D?
The usual cause is the data, not the tool. Search is weak when the library is messy, duplicated, or named inconsistently. Drafts come out generic on the default settings, and checks come out generic when nobody supplies a real checklist.
Is it safe to paste company information into a generative AI tool?
It depends on the tool and on your company's policy. Treat anything you paste into a personal account as external, and check what your organization allows before you share confidential data.
Does generative AI replace the scientist or the reviewer?
No. Author writes a draft for a person to review and finish, and Checker catches basic gaps before a human reviewer sees the document. People still sign off.
Where should an R&D team start with generative AI?
Pick one pain point where information is scattered or documents take a long time to write, such as turning market signals into product ideas, checking drafts against regulatory requirements, writing reports and stage-gate documents, or pulling data out of lab files in different formats. Test it on a past project where you already know the answer, so you can judge the result.
Can generative AI propose new molecules or formulations?
Special models trained on chemical structures can, but that is a different technology from the chat tools in this article. A computational chemistry or materials informatics team usually runs them, rather than someone typing a request into a chat window.