Making the Best Use of AI: Why Plugins, Appsal intelligence is powerful on its own. 

It can generate ideas, analyse information, improve writing, create plans, suggest solutions and help us think through a problem.

However, AI becomes considerably more useful when it is connected to other specialist tools.

What many people still call plugins are increasingly described as apps or connectors. These allow an AI system such as ChatGPT to work alongside services including Canva, cloud storage, calendars, spreadsheets, research tools and business applications. Instead of simply telling us what to do, the AI can help move the work into the application where it will be completed. nges AI from being merely a clever conversational assistant into something closer to a creative director, project coordinator and production assistant.

The difference is not simply speed. Used properly, these connections can improve the quality, consistency and usefulness of the final result.

AI Is Powerful, but It Cannot Do Every Job Equally Well

One of the mistakes people make when using AI is expecting one system to perform every part of a task perfectly.

ChatGPT may be excellent at:

• developing an idea;

• researching a subject;

• organising information;

• writing a detailed creative brief;

• suggesting a visual structure;

• reviewing a draft;

• identifying weaknesses;

• producing revised instructions.

Canva, meanwhile, specialises in:

• creating layouts;

• combining text and graphics;

• applying fonts and colours;

• producing social media graphics;

• preparing presentations;

• resizing designs;

• working with templates;

• producing editable visual content.

The two tools have different strengths.

ChatGPT can decide what the design needs to communicate. Canva can help turn those instructions into an actual visual design.

That is the real value of connecting AI systems to specialist applications.

Think of the AI as the Director

A useful comparison is a film production.

The director does not personally operate every camera, build every set, adjust every light and edit every frame. The director decides what the production is trying to achieve and coordinates the specialists who make it happen.

AI can take a similar role.

It can help determine:

• the audience;

• the objective;

• the central message;

• the visual hierarchy;

• the emotional tone;

• the words that should appear;

• the information that should be omitted;

• the most suitable format.

It can then pass a structured set of instructions to Canva.

This is much more effective than entering a vague instruction such as:

Make me a good social media image about AI.

Canva may still produce something attractive, but “good” is subjective. The tool does not yet know who the image is for, where it will be published, what message matters most or what should receive the viewer’s attention.

The more precise the direction, the more controlled the result.

A Practical Example: Creating a Social Media Image

Suppose I have written a blog called:

Making the Best Use of AI

I want a square image for LinkedIn, Facebook and Instagram.

My first instruction might be:

Create a social media image about using ChatGPT with Canva.

This might generate a reasonable starting point. However, it leaves most of the important creative decisions to the system.

A stronger instruction would be:

Canva, create a 1080 × 1080-pixel social media graphic promoting a business blog called “Making the Best Use of AI”.

Show a professional workspace in which an AI conversation develops into a polished graphic design. Use a clear left-to-right visual journey: an idea or prompt on the left, an AI processing stage in the centre and a finished social media design on the right.

Use a modern, intelligent and practical style rather than science fiction. Avoid robots, glowing humanoid faces, excessive circuitry and meaningless streams of computer code.

Use a dark navy, white and warm gold colour palette. Add subtle blue highlights. Keep the background uncluttered.

The main headline must read: “Making the Best Use of AI”.

Add the smaller supporting line: “Better prompts. Connected tools. Better results.”

Make the headline the strongest visual element. Leave generous margins around all text and keep important content away from the edges.

Add “Philip M Russell Ltd” discreetly at the bottom.

The finished image should look professional, educational and suitable for a business audience.

That instruction controls:

• the dimensions;

• the purpose;

• the audience;

• the composition;

• the style;

• the colours;

• the headline;

• the supporting message;

• the branding;

• the elements to avoid.

The design tool is no longer being asked to guess what I want. It has been given a proper creative brief.

Canva’s ChatGPT app can create, edit and preview designs from within a conversation, with the option to continue customising the result in Canva. irst Result Should Be Treated as a Draft

One of the biggest misunderstandings surrounding AI is the belief that a prompt should produce a perfect result immediately.

Professional creative work rarely operates that way.

A photographer does not expect every frame to be the final photograph. A writer revises a first draft. A designer experiments with layout, spacing and visual hierarchy.

AI-generated work should be approached in the same way.

The first result gives us something concrete to examine.

We can then ask:

• Is the headline easy to read?

• Is the purpose immediately obvious?

• Is the image too busy?

• Does the background compete with the text?

• Does it look like our business?

• Is the supporting text too small?

• Does the image contain unnecessary generic AI imagery?

• Will it still work when viewed on a mobile telephone?

• Is there enough contrast?

• Is the most important element receiving the most attention?

This is where ChatGPT becomes useful again.

Using ChatGPT as an Art Director

After Canva has produced the first design, the result can be reviewed by ChatGPT.

A useful review instruction might be:

Act as an experienced social media art director. Review this design for clarity, visual hierarchy, mobile readability and professional credibility.

Identify the three most important weaknesses. Then write precise editing instructions for Canva. Do not redesign it completely unless necessary. Preserve the central concept and brand colours.

ChatGPT might then recommend changes such as:

Increase the headline size and reduce the amount of decorative background detail.

Move the finished-design panel slightly to the right so that it does not compete with the central AI symbol.

Increase the contrast behind the supporting line and shorten it to improve mobile readability.

Reduce the number of glowing interface elements because they make the design appear too much like a generic technology advertisement.

Those instructions can then be sent back to Canva.

The process becomes a creative loop:

Idea → brief → design → review → correction → final design

This is much more powerful than asking for a single image and accepting whatever appears.

Write Editing Commands, Not Vague Opinions

When improving a design, vague reactions are not particularly useful.

Saying:

I don’t really like it.

does not tell the AI what needs to change.

A controlled editing command should identify:

1. what is wrong;

2. what should change;

3. what must remain unchanged;

4. why the change is needed.

For example:

Keep the existing layout and navy-and-gold colour palette. Remove the robot illustration and replace it with a clean visual showing a written prompt becoming an editable Canva design. Make the image feel practical and businesslike rather than futuristic. Do not change the headline or company name.

Another useful command would be:

Preserve the background image and overall composition. Enlarge the headline by approximately 15 per cent, move it slightly upwards and add a subtle dark overlay behind it. Increase the contrast without covering the main subject.

These are production instructions rather than general comments.

The clearer the correction, the less likely the system is to alter parts of the design that were already working.

A Seven-Part Formula for Better Canva Instructions

When asking ChatGPT to write instructions for Canva, I find it useful to include seven elements.

1. Purpose

What is the image intended to achieve?

Is it promoting a blog, advertising a course, explaining a scientific idea or announcing an event?

2. Format

Specify the required size and destination.

For example:

• 1080 × 1080 pixels for a square social post;

• 1200 × 628 pixels for a blog or link preview;

• 1080 × 1920 pixels for a story;

• A4 portrait for a printable information sheet;

• 16:9 for a presentation slide or YouTube thumbnail.

3. Audience

A design for A-level students should not necessarily look like a design for company directors.

State who needs to understand or respond to the image.

4. Main Message

Decide what the viewer should understand within the first few seconds.

One strong headline is usually better than several competing messages.

5. Visual Direction

Describe the scene, layout, focal point, photographic style or illustration style.

“Professional photograph of a teacher using AI” is still broad.

“Over-the-shoulder photograph of an experienced teacher using an AI assistant to develop a colourful science worksheet, with laboratory equipment softly visible in the background” is far more useful.

6. Brand Controls

Include colours, logo position, company name, preferred style and any fonts or visual conventions that matter.

Where supported, connected design workflows can also apply brand assets and produce editable content rather than leaving the user with an unchangeable flat image. Canva has also introduced technology that can convert generated imagery into layered, editable designs. egative Instructions

Tell the system what must not appear.

For example:

• no robots;

• no distorted hands;

• no meaningless code;

• no excessive neon lighting;

• no tiny text;

• no clutter;

• no American classroom imagery;

• no unrelated icons;

• no invented company logos.

Negative instructions are particularly important when trying to avoid the familiar, generic appearance of AI-generated advertising.

The Importance of Keeping Designs Editable

A completely finished flat image may look impressive, but it can become a problem when something needs to change.

Perhaps the date is wrong. The headline needs shortening. The image needs resizing. A logo must be moved. The same design is required for LinkedIn, Instagram and a website banner.

Editable design elements are therefore extremely valuable.

Text should remain editable. Backgrounds should be replaceable. Logos should be movable. Individual objects should be adjustable. Colours should be changeable.

This allows AI to create the first version without taking control away from the human user.

The objective should not be to remove people from the design process. It should be to remove unnecessary repetitive work while preserving human judgement.

One Design Can Become a Complete Campaign

Once the main image has been approved, ChatGPT can help create instructions for a family of related materials.

For example:

Using the approved square design as the master style, create:

• a 1200 × 628-pixel blog header;

• a 1080 × 1920-pixel story;

• a LinkedIn banner;

• a presentation title slide;

• an A4 promotional poster.

Preserve the navy, white and gold palette, the central visual concept and the company branding. Adjust the layout for each format rather than simply stretching the original. Keep all text editable.

This is where the combination begins to save substantial time.

The AI is no longer creating one isolated graphic. It is helping establish a consistent visual campaign.

The same approach can be used for:

• course advertisements;

• sailing event announcements;

• science experiment videos;

• company blog promotions;

• educational worksheets;

• YouTube thumbnails;

• presentation slides;

• posters;

• photographic exhibitions;

• product information.

AI Can Also Help Before Canva Is Opened

The connection between ChatGPT and Canva is useful, but the preparation can begin even earlier.

ChatGPT can first help decide:

• which words belong on the image;

• which words belong in the accompanying post;

• which visual metaphor would communicate the idea;

• whether a photograph, diagram or illustration would work best;

• which image dimensions are needed;

• how the same campaign should vary between platforms;

• which details are essential;

• which details would create clutter.

For example, I might ask:

Give me five visual concepts for an article about connecting AI to specialist tools. Avoid robots, human brains and glowing circuit boards. Each concept should be understandable without a long explanation.

ChatGPT could suggest:

• an orchestra conductor coordinating specialist instruments;

• a central control desk connected to different creative tools;

• a rough sketch passing through several specialist workstations;

• a workshop in which the AI supplies plans while specialist machines build the result;

• a relay race in which each application completes the stage it performs best.

I can then select the strongest idea before asking Canva to produce anything.

That prevents time being wasted developing the wrong concept.

Connected AI Requires Sensible Judgement

Apps and connectors can make workflows more capable, but they should be used carefully.

Before enabling a connection, it is sensible to understand what information the application may receive and what actions it can perform. OpenAI explains that an enabled app may use relevant conversational context to help fulfil a request. ns we should avoid casually including confidential client information, private student details, passwords, financial data or other sensitive material in a design workflow.

Every final design should also be checked by a person.

AI can:

• misspell words;

• invent details;

• misunderstand branding;

• produce inappropriate imagery;

• create poor contrast;

• position text outside safe areas;

• use an unsuitable visual style;

• alter details that should have remained fixed.

Connection does not remove the need for inspection. It makes informed human supervision more important.

My Own View: The Skill Is Becoming Direction

The more I use AI, the more I realise that the most valuable ability is not simply knowing how to type a prompt.

The real skill is learning how to direct the process.

That means being able to say:

• what the work is for;

• who it is for;

• what success looks like;

• what must remain consistent;

• what needs changing;

• what should be removed;

• when the result is good enough to publish.

A vague prompt transfers nearly every decision to the machine.

A detailed prompt keeps the human in control.

This is particularly important in my own work because I may move between teaching, photography, scientific demonstrations, sailing, video production, music and business administration. The same visual style will not suit every project.

An A-level physics graphic needs clarity and scientific credibility. A sailing photograph needs movement and excitement. A company announcement needs professionalism. A YouTube thumbnail needs immediate impact.

AI can help with all of them, but only when it is given suitable direction.

Conclusion: Connect the Intelligence to the Right Tool

AI is impressive when it can produce an answer.

It becomes far more useful when it can help us complete a process.

Connecting ChatGPT to Canva illustrates the wider opportunity. ChatGPT can develop the idea, write the brief, create the words, suggest the composition and review the result. Canva can produce the layout, manage the visual elements and turn those instructions into an editable design.

Neither tool replaces the other.

More importantly, neither replaces the person directing the project.

The best results come from combining:

human purpose, AI reasoning and specialist applications.

The future of productive AI is not one machine trying to do everything. It is a collection of capable tools, connected intelligently and directed by someone who knows what the finished work needs to achieve.

The first prompt may create something interesting.

The complete, controlled workflow can create something genuinely worth publishing.

 

Which AI Should I Use? Choosing the Right Tool for the Job

Artificial intelligence is no longer a single product or a single website. It is becoming an entire toolbox.

That creates a new problem.

We are constantly being told that we should use AI, but we are not always told which AI we should use for a particular job.

Should we use ChatGPT, Claude, Gemini, Microsoft Copilot or Perplexity? Should we create an image in Canva, Adobe Firefly or ChatGPT? Should we use a specialist system for video, music, coding or voice generation? Should we configure our own AI assistant, or connect several systems together using plugins?

The answer is not to find one AI that supposedly does everything.

The better approach is to understand the job, select the right tool and build a reliable process around it.

There Is No Single “Best AI”

Asking which AI is best is rather like asking:

Which is the best tool in a workshop?

A screwdriver is excellent when you need to turn a screw, but fairly useless when you need to cut a sheet of plywood. A laser cutter is extremely powerful, but it would be ridiculous to use it simply to open a cardboard box.

AI tools work in much the same way.

There is considerable overlap between systems, but they have different strengths, integrations and working methods.

Some are general-purpose assistants. Some specialise in research. Others are built into office software. Some generate images, video, voices or music. Others help programmers write and test code.

At Philip M Russell Ltd, the range of possible AI tasks is already enormous.

On one day, AI might help me:

• turn a set of rough notes into a structured company blog;

• create an X post and a LinkedIn post;

• design a worksheet for an individual student;

• analyse a spreadsheet;

• suggest categories for bank transactions;

• write part of a sailing weather application;

• develop a script for a science video;

• plan an infographic;

• or explore ideas for a piece of music.

Those are all “AI tasks”, but they are not necessarily jobs for the same AI.

Start With the Job, Not the Brand

Before choosing an AI, ask five practical questions.

1. What am I actually trying to produce?

Do you need:

• an idea;

• a polished article;

• a researched answer;

• a spreadsheet;

• an image;

• a video;

• computer code;

• a voiceover;

• a piece of music;

• or an action completed in another system?

The final output often determines which AI is most suitable.

2. Does the answer need current information?

There is a major difference between asking an AI to improve a paragraph you have written and asking it to explain the latest changes to tax rules, examination specifications or energy tariffs.

For current information, you need an AI that can research live sources and show where its information came from.

3. Do I need the AI to use my own material?

An AI may need access to:

• company documents;

• course notes;

• spreadsheets;

• previous articles;

• brand guidelines;

• photographs;

• email messages;

• calendars;

• or technical manuals.

In that case, a Project, knowledge base or connected application may be more useful than an ordinary one-off chat.

4. Does the AI only need to advise me, or must it take action?

There is a substantial difference between:

“Write an email reminding a customer about an invoice.”

and:

“Find the correct customer, create the email, attach the invoice and place the message in my Gmail drafts.”

The first is a writing task.

The second is a connected workflow involving data, permissions and external actions.

5. How sensitive is the information?

Before uploading any document or connecting an account, consider whether it contains:

• personal information;

• student data;

• financial records;

• passwords;

• confidential business material;

• health information;

• or unpublished intellectual property.

Convenience should never replace sensible data handling.

A Practical Guide to Different AI Tools

The following is not a league table. It is a practical guide to where different systems may fit.

Features and subscription arrangements change frequently, so the important lesson is to understand the category of tool rather than memorising a fixed list.

ChatGPT: A Strong General-Purpose Starting Point

ChatGPT is useful for a wide range of tasks, including:

• brainstorming;

• writing and rewriting;

• lesson planning;

• explaining concepts;

• analysing documents;

• working with spreadsheets;

• generating and editing images;

• writing code;

• creating reusable files;

• and building customised assistants.

ChatGPT Projects allow related chats, reference files and custom instructions to be kept together, making them particularly useful for repeated work such as regular content creation, research or planning.

ChatGPT can also analyse common spreadsheet formats, including Excel and CSV files, while its image tools can create or edit images from natural-language instructions.

A good choice for: general business work, teaching resources, planning, writing, data analysis and workflows that combine several different types of task.

Claude: Long-Form Thinking, Documents and Interactive Artifacts

Claude is often useful when working with lengthy documents, detailed analysis, structured writing or substantial pieces of content.

Its Projects can hold their own instructions, documents, knowledge and chat histories. Its Artifacts feature can place substantial content, applications, visualisations and other reusable outputs into a separate working area.

A good choice for: long reports, detailed editing, document analysis, structured thinking and creating interactive prototypes or tools.

Gemini: Google-Connected and Multimodal Work

Gemini is particularly relevant to people who already work extensively with Google services.

Its current feature set includes connected applications, Deep Research, image tools, Canvas, live conversation and custom Gems. Feature access and usage limits vary between subscription levels.

A Gem performs a similar broad role to a customised assistant: it can be configured for a repeatable type of task.

A good choice for: people working heavily with Google services, multimodal questions, web research and Google-centred workflows.

Microsoft 365 Copilot: Working Inside Office Applications

Microsoft 365 Copilot is designed to work within applications such as Word, Excel, PowerPoint, Outlook and Teams.

For example, it can assist with drafting in Word, suggest formulas or analyse data in Excel, summarise email discussions in Outlook and help summarise meetings in Teams.

This can be more convenient than transferring material between an office application and a separate AI chat.

A good choice for: businesses whose documents, spreadsheets, presentations, emails and meetings already live in Microsoft 365.

Perplexity: Current Web Research With Sources

Perplexity describes itself as an AI answer engine that researches the live web and produces concise answers with citations.

It can therefore be useful at the beginning of a project when you need to establish:

• what has recently changed;

• what different sources are saying;

• what research is available;

• or where a particular claim originated.

However, cited research still needs to be checked. A citation proves that a source exists; it does not automatically prove that the source is reliable or that the AI has interpreted it correctly.

A good choice for: initial research, current information and finding sources for further investigation.

Gemini Notebook, Formerly NotebookLM: Learning From Your Own Sources

Gemini Notebook is especially useful when the priority is to work from a defined collection of source material.

You can upload documents and use them to generate summaries, questions, flashcards, quizzes, study materials and audio-style overviews. Its responses are designed to remain grounded in the sources supplied to it.

For a student, this might mean uploading course notes, a specification and several textbook chapters.

For a business, it could mean uploading policies, meeting papers or product documents.

A good choice for: revision, source-based research, document collections and projects where the AI should remain focused on supplied material.

GitHub Copilot: Writing and Maintaining Code

GitHub Copilot is designed specifically for software development.

Its features include code completion, suggestions, code review and agent-style tools that can examine several files, propose changes, run tests and validate results.

A general AI can explain programming and write code, but a dedicated coding assistant may fit more naturally into the development environment.

A good choice for: programmers working regularly with repositories, development environments, testing and code review.

Canva AI: Rapid Social Media and Business Design

Canva AI combines design, writing, image and video tools within the Canva environment.

Its Magic Media tools can generate images, graphics and videos from descriptions, while Magic Design can help turn supplied material into layouts and short videos.

For a small business already creating posters, social graphics, presentations and promotional material in Canva, this can remove several unnecessary stages.

A good choice for: social graphics, posters, presentations, branded content and rapid design work.

Adobe Firefly: Creative Production and Adobe Workflows

Adobe Firefly generates and edits images, video, audio, vectors and designs. It also integrates with the wider Adobe Creative Cloud environment.

This is particularly relevant when AI-generated material needs to move into applications such as Photoshop, Premiere Pro or other professional creative tools.

A good choice for: photographers, designers, editors and video producers already using Adobe software.

Runway: AI-Assisted Video Generation

Runway concentrates heavily on generative video.

It can generate, edit and extend video using text, images and existing footage.

This might be used for:

• concept sequences;

• illustrative B-roll;

• visual experimentation;

• storyboarding;

• background material;

• or creating shots that would otherwise be difficult to film.

A good choice for: dedicated AI video generation and experimentation.

ElevenLabs: Voice Generation and Narration

ElevenLabs specialises in synthetic speech, voice design, voice cloning, transcription and voice-based agents.

Its text-to-speech tools are aimed at uses such as video narration, games, podcasts, audiobooks and accessibility.

Naturally, voice cloning must only be used with proper permission. The fact that technology can imitate a voice does not mean that it is acceptable to imitate anyone without their knowledge.

A good choice for: narration, accessibility, multilingual audio and approved synthetic voices.

Suno: Generating Musical Ideas and Complete Songs

Suno is designed to turn a written idea, recorded sound, rhythm or melody into a more complete musical production.

It can generate lyrics, beats, vocals and full song arrangements.

For a musician, it can provide a useful starting point, but the most interesting results may come from taking those ideas into a digital audio workstation and developing them further.

A good choice for: musical experimentation, song concepts, backing ideas and rapid demonstrations.

You Do Not Have to Accept the AI’s Default Behaviour

One of the most powerful developments is the ability to configure a general AI for a particular job.

Instead of repeatedly explaining your company, preferred tone and required output, you can create a dedicated Project or customised GPT.

A custom GPT can combine instructions, uploaded knowledge, selected capabilities, connected apps or defined API actions.

This turns a general-purpose AI into something much closer to a task-driven assistant.

Example: Creating a Philip M Russell Ltd Content Assistant

Suppose I want an AI whose job is to create the daily Philip M Russell Ltd blog and supporting social content.

Step 1: Create the Workspace

I could create a ChatGPT Project called:

Philip M Russell Ltd Daily Blog and Social Media

Alternatively, I could create a custom GPT called:

PMR Content Assistant

A Project is useful when I want to keep an evolving collection of chats and material together.

A custom GPT is useful when I want a clearly defined assistant that performs the same role whenever it is opened.

Step 2: Add Relevant Knowledge

I might upload:

• a description of Philip M Russell Ltd;

• details of the tuition services;

• examples of previous blogs;

• the preferred UK English writing style;

• company biographies;

• lists of science equipment;

• sailing projects;

• photography and video equipment;

• preferred hashtags;

• and examples of successful social posts.

The assistant would then have a far better understanding of the company than an AI beginning with a blank conversation.

Step 3: Give It Precise Instructions

The following could be placed in the Project instructions or custom GPT configuration:

Role
You are the content assistant for Philip M Russell Ltd, a UK education, science, technology and media company.

Primary task
Turn the owner’s rough notes into a detailed, engaging company blog.

For every topic, produce:

1. A strong title.

2. An engaging introduction.

3. A detailed structured blog using clear section headings.

4. Practical examples.

5. Relevant personal reflections written in the first person.

6. A compelling conclusion.

7. A short curiosity-driven X post.

8. A thoughtful LinkedIn post that builds professional authority.

9. Strong but relevant hashtags.

10. Three detailed Canva image prompts.

Writing rules

• Use UK English spelling.

• Keep the tone knowledgeable, reflective and accessible.

• Do not invent events, qualifications, results or customer claims.

• Clearly distinguish fact from personal opinion.

• Avoid exaggerated marketing language.

• Explain technical ideas in plain English.

• Where current factual information is required, research and cite reliable sources.

Quality check before finishing
Confirm that the article has a clear central argument, practical value, a personal voice and a conclusion that adds something new rather than merely repeating the introduction.

That instruction set is much more reliable than simply saying:

“Write me a blog about AI.”

Step 4: Use a Consistent Input Form

Each day, I could provide:

Topic: Making a pennant for Champagne
What happened: We began planning a traditional pennant for the Thames A-Rater.
Important details: Shape, colours, logo, fabric, stitching, visibility and how it will appear in photographs.
Personal angle: A pennant is a small item, but it gives the boat identity and character.
Audience: Sailing enthusiasts, makers and small-business readers.

The AI now has a clear task, reliable background information and an agreed output format.

This is the real value of a configured AI.

It does not necessarily make the AI more intelligent. It makes the process more disciplined and repeatable.

What Is a Plugin?

A plugin is easiest to understand by separating three things.

The AI Is the Brain

The AI interprets the request, reasons about the information and decides what needs to happen.

The Connected App Provides Eyes and Ears

A connected app might allow the AI to retrieve information from:

• email;

• cloud storage;

• a calendar;

• a customer database;

• a spreadsheet;

• or another approved service.

An Action Provides Hands

An action may allow the AI to:

• create a draft;

• update a record;

• add a calendar event;

• create a task;

• write information into another system;

• or call a company’s own API.

In current ChatGPT terminology, plugins provide a way to discover and enable workflow capabilities. A plugin may contain skills, app connections and templates. The underlying apps connect ChatGPT to external information and actions.

Some connected apps can search information, synchronise content, support research or carry out approved write actions. Availability depends on the plan, application, permissions and workspace settings.

How a Plugin-Based Process Works

A connected workflow usually follows a sequence.

1. The User Connects a Service

The user selects an approved plugin or app and signs into the relevant service.

This should be treated as granting access, not merely installing a decorative extra.

2. Permissions Are Defined

The connection may be allowed to:

• read information;

• search information;

• create new items;

• modify existing items;

• or perform only a limited set of actions.

Good practice is to grant only the access required for the job.

3. The User Gives the AI a Defined Task

For example:

“Find my notes about this week’s science demonstrations, create a blog, prepare two social posts and place the finished email newsletter in my drafts.”

4. The AI Retrieves the Necessary Information

The connected app locates the approved files, messages or records.

5. The AI Processes the Material

The AI might:

• summarise;

• classify;

• compare;

• calculate;

• draft;

• or transform the information.

6. The System Requests Approval Where Necessary

Actions with a meaningful external effect should be checked before they are completed.

Creating an email draft is not the same as sending it.

Suggesting a spreadsheet category is not the same as submitting accounts.

Preparing a calendar event is not the same as inviting fifty people to it.

7. The Action Is Completed

Once approved, the connected system creates or updates the relevant item.

8. A Human Checks the Result

The process should end with verification, particularly where the action affects customers, students, money or public communications.

Practical Plugin Workflow Examples

A Content Production Workflow

A connected content assistant could:

1. Read upcoming events from a calendar.

2. Find associated notes in cloud storage.

3. identify the photographs available for the topic.

4. Draft the blog.

5. Create platform-specific social posts.

6. Produce image instructions.

7. Create an email newsletter draft.

8. Leave everything ready for human review.

The final publication decision remains with the business owner.

A Bookkeeping Workflow

A bookkeeping process could:

1. Retrieve a bank statement.

2. Transfer the transactions into a spreadsheet.

3. Apply known categories to familiar suppliers.

4. Flag uncertain items.

5. identify possible duplicates.

6. Produce a list of missing receipts.

7. Create reminders for unresolved transactions.

The AI should not pretend to know what an ambiguous transaction represents. A good system highlights uncertainty rather than hiding it.

A Tuition Workflow

A tuition assistant could:

1. Read the lesson topic from the calendar.

2. Retrieve the relevant course specification.

3. Find the student’s previous focus sheet.

4. Create a personalised worksheet.

5. Produce easier and harder versions.

6. Draft a parent progress summary.

7. Save the material for review before the lesson.

That is much more useful than producing the same generic worksheet for every student.

A Sailing Information Workflow

A connected sailing assistant could:

1. retrieve wind and weather information;

2. collect river level and flow data;

3. check for relevant warnings;

4. produce a club-focused summary;

5. generate a consistent weather graphic;

6. and prepare the information for publication.

Where a company has its own data service, a custom GPT action can be configured to use an external API. This requires the API’s authentication details and a schema defining the available endpoints and operations.

Plugins Do Not Remove Responsibility

Connecting more systems does not automatically make a process better.

It may simply allow a mistake to travel further and faster.

A poorly designed automated workflow might:

• use the wrong source document;

• send an unfinished message;

• expose private information;

• overwrite a correct record;

• misunderstand a customer request;

• or repeat an error across hundreds of entries.

OpenAI’s own guidance warns that connected agents may encounter prompt-injection attacks or gain access to sensitive information. It recommends enabling only the apps required for a task, avoiding vague instructions and reviewing permissions regularly.

A sensible AI process therefore needs:

• limited permissions;

• clear instructions;

• reliable source data;

• approval points;

• exception handling;

• and human verification.

The Best Solution May Be an AI Team

Sometimes the most effective solution is not one AI but a sequence of specialist tools.

A content workflow might look like this:

1. Use a research-focused AI to locate current sources.

2. Use ChatGPT or Claude to organise the argument.

3. Use a configured assistant to apply the company’s style.

4. Use Canva or Firefly to produce the visual material.

5. Use ElevenLabs for an approved narration.

6. Use a human editor to check the final result.

7. Use a connected app to prepare the content for publication.

Each AI performs the part of the job for which it is most suitable.

The human remains responsible for the complete result.

Do Not Automate a Bad Process

Before adding AI, it is worth asking whether the existing process makes sense.

If a task is disorganised, poorly defined or based on inaccurate information, automation will not repair it automatically.

It may simply perform the bad process more quickly.

The best order is:

1. Understand the task.

2. Remove unnecessary stages.

3. Standardise the information.

4. Decide where human judgement is required.

5. Select the appropriate AI.

6. Add connections or automation carefully.

7. Measure whether the result has genuinely improved.

Conclusion: Build an AI Toolbox, Not an AI Dependence

The question is not:

“Which AI is the best?”

The better question is:

“Which AI is best suited to this particular stage of this particular job?”

A general assistant may help us think and write.

A research AI may help us find current sources.

A specialist image, video, voice or music system may produce the creative asset.

A configured Project or custom GPT may make a repeated task more consistent.

A plugin may connect the AI to the information and applications needed to complete a process.

The important skill is no longer merely knowing how to ask an AI a clever question.

It is understanding how to design the whole workflow:

• choosing the right tool;

• supplying the right information;

• defining the task;

• controlling permissions;

• checking uncertainty;

• and keeping human judgement at the centre.

AI should not become an excuse to stop thinking.

Used properly, it should allow us to spend more time thinking about the parts of the job that genuinely matter.

So How do I produce blogs like this?

I generate the ideas.  AI's simply go off and do their own thing - not related to what I do.

Voice to document - I dictate much of the story. I want this blog to be about me not an AI. Auto spell correct helps here though.

Then I can get an AI to clean up the text - suggest better orders for paragraphs and or ideas.

Then the magic bit for me - the AI can put in headings in the correct places - this really is the AI in action, then either sort through my vast library of photographs - you really don't want to know how many - It's in the many terabyte order - and usual manually adding text or generate my own.

Practice at writing suitable prompts for an Ai graphic engine and many attempts can produce an image I don't have or it can cobble several i do have together.

 

Playing Catch-Up with AI: Why Teachers and Employers Cannot Afford to Stand Still

Like it or not, artificial intelligence is already with us.

It is not something waiting in the distance. It is not a technology that might eventually affect education, employment or business. It is already being used every day by students, teachers, employees, managers and customers.

The uncomfortable truth is that many students are probably using AI more confidently than the adults teaching them.

They know how to ask it questions, generate ideas, produce graphics, organise information and rewrite text. Some can ask AI to produce an essay and then repeatedly alter the language until it appears more like their usual work. Others use it to make impressive maps, diagrams, presentations, revision cards and infographics.

This creates obvious problems around authenticity and assessment. However, it also points to a much wider issue.

While we are discussing whether students should be allowed to use AI, many of them have already moved on to deciding how they are going to use it.

Teachers, schools, businesses and other organisations are now playing catch-up.

AI Has Already Entered the Classroom

There is a temptation to think that banning AI will solve the problem.

A school might block a particular website. A teacher might warn students that AI-generated homework will be detected. An organisation might insist that employees complete every task without AI assistance.

Yet AI is increasingly built into search engines, word processors, design platforms, mobile phones, coding environments and office software. Even when one system is blocked, another is usually available.

Students can use AI to:

• explain difficult concepts;

• generate essay plans;

• summarise articles;

• create revision questions;

• check calculations;

• produce computer code;

• design presentations;

• improve grammar;

• generate images;

• create maps and diagrams;

• turn notes into flashcards;

• rehearse examination questions.

Some of these uses may undermine learning. Others can enhance it enormously.

The real question is no longer simply, “How do we stop students using AI?”

A better question is:

How do we teach students to use AI without allowing it to replace their thinking?

The Problem Is Not AI — It Is Uncritical Use

A calculator can help someone perform a difficult calculation, but it cannot decide whether the calculation is appropriate.

A spellchecker can identify a misspelled word, but it cannot always tell whether the chosen word makes sense.

AI works in much the same way. It can produce fluent, convincing answers, but those answers may be incomplete, biased, outdated or simply wrong.

The danger is not merely that students will use AI. The greater danger is that they will trust it without checking it.

A student might ask an AI system to explain a scientific process and receive a confident but inaccurate answer. Another might submit an essay containing invented quotations or references. A mathematics student might be shown a method that looks plausible but includes a hidden algebraic error.

Using AI well therefore requires several important skills:

• asking clear questions;

• checking sources;

• testing calculations;

• identifying assumptions;

• recognising uncertainty;

• comparing different explanations;

• editing rather than merely accepting;

• taking responsibility for the final work.

These are not shortcuts around education. They are increasingly part of education.

Why Teachers Need to Catch Up

Many teachers are already under enormous pressure. They have lessons to plan, work to mark, reports to write, meetings to attend and administrative systems to maintain.

Learning another technology may feel like one more demand on an already overcrowded timetable.

However, teachers do not need to become computer scientists. They need enough confidence to understand what AI can do, what it cannot do and where it might genuinely help.

The starting point does not have to be complicated.

A teacher could begin by asking AI to:

• suggest several ways of introducing a difficult topic;

• turn a lesson objective into a sequence of activities;

• create ten retrieval questions;

• adapt a worksheet for different levels of ability;

• generate example examination answers;

• produce misconceptions for students to correct;

• create a vocabulary list;

• draft a parent communication;

• suggest practical demonstrations;

• organise existing notes into a clearer structure.

The teacher must still inspect, correct and adapt the result. AI should not be treated as an unquestionable authority.

Used sensibly, however, it can provide a useful first draft and reduce the time spent staring at a blank page.

Moving Beyond Generic Worksheets

One of the most powerful educational uses of AI is personalisation.

Traditional worksheets are normally designed for a whole class. They may be suitable for many students, but not necessarily for all of them.

One student may need more basic practice. Another may understand the topic but make careless arithmetic errors. A third may need extension questions. Someone else may require shorter instructions, more diagrams or additional scaffolding.

Producing separate resources for every student by hand would be extremely time-consuming.

AI can help create variations much more quickly.

Imagine a class learning quadratic equations. A teacher could create:

• a supported worksheet with worked examples;

• a standard worksheet covering the core method;

• a worksheet focusing on common sign errors;

• an extension sheet involving unfamiliar contexts;

• a short confidence-building exercise for a struggling student;

• a challenge set for a student preparing for a top grade.

The learning objective remains the same, but the route towards it becomes more appropriate for each learner.

This does not remove the teacher. It makes the teacher’s professional judgement more powerful.

Personalisation in Private Tuition

In one-to-one tuition, individualised learning has always been important.

Two students studying the same A-level Mathematics topic may have completely different needs. One may understand differentiation but fail to show sufficient working. Another may know the method but struggle with algebra. A third may rush the question and misread an instruction.

A generic collection of twenty questions may not address any of these problems effectively.

AI can help produce a carefully targeted progression.

For example, after noticing that a student repeatedly makes mistakes when applying the product rule, I could ask for:

1. three simple questions concentrating on identifying the two functions;

2. three questions requiring the product rule with clear scaffolding;

3. three questions involving algebraic simplification;

4. two examination-style problems where the method is not stated;

5. one final challenge question combining the product and chain rules.

I would still need to check every question and solution. I would also need to observe how the student responds.

The AI creates material. The teacher diagnoses the learning.

That distinction is crucial.

AI Can Help Make Science More Visible

AI should not be limited to producing text.

In science teaching, it can help generate diagrams, experimental instructions, risk-assessment drafts, data tables, graphical examples and questions based on practical observations.

Suppose I am teaching electric fields using a Van de Graaff generator.

AI might help me create:

• a labelled diagram of the apparatus;

• a prediction sheet;

• questions about the motion of a charged object;

• a comparison between electric and gravitational fields;

• a set of common misconceptions;

• extension questions about potential difference;

• a simplified explanation for a younger student;

• an examination-style six-mark question.

The experiment itself remains central. Students still need to observe, measure, explain and evaluate.

AI does not replace the flash of the spark, the movement of the pith ball or the unexpected behaviour that prompts a real scientific question.

Instead, it can help build stronger learning around the practical experience.

Turning Students from Consumers into Critics

Perhaps the most valuable classroom activity is not asking AI to provide the correct answer.

It is asking AI to provide an answer that students must evaluate.

A teacher could generate a deliberately imperfect explanation and ask students to identify the problems.

For example:

“A heavier object falls faster because gravity pulls on it more strongly.”

Students could be asked:

• What part of this statement is true?

• What important idea is missing?

• How does mass affect gravitational force?

• Why do objects have the same acceleration in a vacuum?

• How could this be tested experimentally?

This changes the student’s role. They are no longer simply receiving an answer. They are checking, challenging and improving it.

The same technique can be used in other subjects.

In English, students can improve a weak paragraph.

In history, they can identify unsupported claims.

In geography, they can assess an oversimplified explanation of migration.

In psychology, they can check whether a study has been represented accurately.

In computing, they can debug generated code.

In mathematics, they can locate the first incorrect line in a proposed solution.

AI becomes material for thinking rather than a substitute for thinking.

Assessment Will Have to Change

AI exposes a weakness that has existed in education for a long time.

If a piece of homework can be completed successfully by copying information from a textbook, downloading an answer or asking an AI system to write it, perhaps the task was not measuring deep understanding in the first place.

This does not mean written homework is useless. It means we need to think more carefully about what the task is designed to reveal.

More robust assessment might include:

• asking students to explain their reasoning verbally;

• requiring drafts and evidence of development;

• discussing why particular sources were chosen;

• using local or personal examples;

• completing part of the task under supervision;

• evaluating an AI-generated answer;

• comparing alternative methods;

• reflecting on mistakes;

• applying knowledge to an unfamiliar situation;

• demonstrating a practical skill.

A student who understands their work should be able to discuss it, defend it, modify it and apply it.

A student who has merely submitted generated text will often struggle to do those things.

The solution is not an endless technological contest between AI generation and AI detection. It is better task design, clearer expectations and more meaningful assessment.

AI Detection Is Not a Complete Solution

It is easy to imagine a simple contest.

Students use AI to produce work. Teachers use another AI system to detect it.

Unfortunately, this can create false confidence. AI-detection systems may misclassify genuine writing, while edited or mixed-origin work may be difficult to identify reliably.

There are also important questions of fairness. A student should not be accused of misconduct solely because a piece of software produces a probability score.

Teachers still need evidence, context and professional judgement.

Changes in vocabulary, sudden shifts in quality, invented references, inability to explain the work and inconsistency with supervised performance may all raise legitimate questions. However, these need to be investigated carefully rather than treated as automatic proof.

The best long-term response is to create a culture in which students are expected to disclose how AI was used.

For example:

“I used AI to generate possible essay headings. I selected three, changed the order and wrote the final argument myself.”

Or:

“I asked AI to explain this calculation in two different ways. I checked the method against my notes and then completed the questions independently.”

This makes the process visible and encourages responsible use.

AI Can Reduce Repetition — But That Is Only the Beginning

There are many repetitive tasks that AI can help with.

In education, these may include:

• producing first drafts of lesson plans;

• creating worksheets;

• generating question banks;

• drafting routine communications;

• formatting notes;

• adapting reading levels;

• creating mark-scheme outlines;

• summarising meeting notes;

• converting material into quizzes;

• organising revision schedules.

In business, AI can help with:

• categorising transactions;

• drafting standard emails;

• summarising documents;

• organising customer enquiries;

• preparing reports;

• producing marketing ideas;

• rewriting technical information for different audiences;

• creating social media content;

• extracting actions from meeting notes.

These uses matter because repeated administrative work consumes time and attention.

However, the greatest value of AI may not be saving time. It may be increasing what a person can accomplish during that time.

A teacher might use the saved time to provide better feedback.

A designer might explore five concepts instead of one.

A business owner might analyse patterns that were previously overlooked.

A tutor might produce resources targeted to the precise errors made by an individual student.

Productivity is not simply doing the same work faster. It is being able to produce better work, explore more possibilities and respond more effectively.

AI May Not Shorten the Working Day

There is a popular claim that AI will save everyone hours of work.

Sometimes it will.

However, new technology often raises expectations as well as reducing effort. When a task becomes faster, people frequently produce more versions, offer more personalisation, respond more quickly or take on additional work.

AI may not necessarily mean that teachers finish several hours earlier.

Instead, it might mean that they can:

• produce more differentiated resources;

• create better revision materials;

• give more detailed feedback;

• communicate more clearly with parents;

• explore additional teaching strategies;

• support a wider range of learners.

That is still an important benefit.

The purpose of AI should not be to turn teachers into faster administrative machines. It should be to create more room for judgement, explanation, creativity and human interaction.

Knowing Where to Start

For someone new to AI, the range of possibilities can feel overwhelming.

The simplest approach is to begin with one real task.

Do not start by asking, “How can AI transform everything I do?”

Start with:

“What repetitive task takes too much of my time?”

Then try using AI to create a first draft.

A teacher might begin with a worksheet.

A business owner might begin with a routine customer email.

A student might begin with a revision timetable.

After receiving the result, ask:

• Is this accurate?

• Is it suitable for the intended audience?

• What needs changing?

• What has been omitted?

• Could the instructions be clearer?

• Does the tone sound appropriate?

• Would I be confident putting my name to it?

If the first result is poor, that does not necessarily mean the tool is useless. The request may need more context.

Instead of asking:

“Make me a worksheet on forces.”

Try:

“Create a 30-minute GCSE Physics worksheet on resultant forces for a student working towards Grade 7. Begin with three recall questions, include four calculation questions using force diagrams, add one misconception question and finish with a six-mark examination-style problem. Provide a separate answer sheet.”

The quality of the instruction usually has a major influence on the quality of the result.

AI can even help improve the instruction. A user can describe what they are trying to achieve and ask the system what additional information it needs.

The Teacher Remains Essential

AI can generate explanations, but it cannot fully understand the student sitting in front of me.

It may not notice hesitation before an answer.

It may not recognise when a student is pretending to understand.

It cannot always tell whether a mistake comes from weak subject knowledge, anxiety, poor reading, rushed work or a lack of confidence.

It does not know when to abandon the planned lesson because a more important misconception has appeared.

It cannot replace the encouragement given when a student finally understands something they previously thought was impossible.

Teaching is not simply the delivery of information.

It is diagnosis, communication, motivation, adaptation and human judgement.

AI can support those things, but it should not be allowed to erase them.

From Resistance to Responsible Adoption

We should not pretend there are no risks.

AI can encourage plagiarism, reduce independent thinking, produce misinformation and create impressive-looking work without genuine understanding. It also raises questions about privacy, copyright, bias, data protection and unequal access.

These issues need clear rules and serious discussion.

However, refusing to engage with AI does not make those problems disappear. It simply leaves students and employees to work them out without guidance.

Education should prepare people for the world they are entering, not the world we remember.

That means teaching people:

• when AI is useful;

• when it is inappropriate;

• how to acknowledge its use;

• how to verify its output;

• how to protect confidential information;

• how to retain ownership of their thinking;

• how to recognise when human judgement matters more.

Conclusion: The Real Risk Is Standing Still

AI is not going away.

Students are already experimenting with it. Businesses are already adopting it. Software companies are building it into the tools we use every day.

The choice is not between a world with AI and a world without it.

The choice is between using it carelessly and learning to use it intelligently.

Teachers do not need to compete with their students in every new application. They do, however, need enough knowledge to guide them. Employers do not need to automate every task, but they should understand where AI could improve productivity, creativity and decision-making.

Most importantly, we must stop thinking of AI only as a machine for producing finished answers.

Its greater educational value lies in helping us ask better questions, explore alternatives, identify misconceptions, personalise learning and extend what teachers and students can achieve.

There will still be mistakes. There will still be misuse. There will still be situations where the best decision is not to use AI at all.

But standing still is not a neutral choice.

The organisations that learn how to combine artificial intelligence with human expertise will be better prepared than those that simply hope it will disappear.

AI can produce text, graphics, calculations and resources.

What it cannot decide is what kind of education we want to create.

That responsibility remains ours.