Productivity

Don't Buy Ten AI Subscriptions: A Stack for Engineering and Admin

By Christhian Chassi · Developer & Product Manager · Published August 17, 2026 · 12 min read

For most of my career as a project manager, the toolkit was three things: Word, Excel and Microsoft Project. That was the job. It stopped being enough.

What changed isn't that those tools got worse — it's that the range of business areas I'm expected to cover got wider. Engineering questions, financial modelling, operational dashboards, technical documentation, project governance. The same person is now expected to front all of it, and Word and Excel alone don't carry that load.

So I rebuilt the toolkit around AI tools. What follows is what I'd actually buy, in what order, and the two workflows they fit into. Not thirteen reviews — a stack.

The mistake to avoid first

Don't buy ten subscriptions. This is the single most expensive error in this category, and it's easy to make because every tool demos well in isolation.

Most of these products overlap heavily. Paying for four things that all do competent general reasoning gets you one capability at four times the price, and a tab problem. The stack below is deliberately small: one primary assistant, one specialist per genuinely distinct job.

Why nobody serious runs just one

There's a tension in what I just said, so let me resolve it: don't buy ten is not the same as buy one.

Almost everyone I know who uses AI intensively ends up running two to four tools and switching by task. That isn't indecision, and it isn't a failure to commit. It's the same instinct that stops you doing financial modelling in Word — different jobs have different right answers, and pretending otherwise costs you quality.

My own count is higher than average because my work spans more ground: at least five tools for engineering work, and at least four for growth and design. Not five subscriptions I felt like buying — five because the engineering chain genuinely has five distinct steps, and no single product covers them.

The failure mode isn't owning several tools. It's owning several tools that do the same thing. One versatile primary, then one specialist per genuinely distinct job, is a stack. Four general assistants is just an expensive habit.

The six I'd actually pay for, in order

If you're covering engineering plus administration, this is the sequence I'd buy in. Stop wherever your work stops needing the next one — there's no prize for owning all six.

• 1. ChatGPT — your primary. The most versatile of the set: data analysis, Excel work, programming, documents, calculations, brainstorming, automation. If you buy one thing, buy this. Its weakness is worth knowing: it can get calculations and technical specifics wrong, and it doesn't replace specialist software. Verify anything numeric that matters.
• 2. Claude — for documents, reasoning and Excel. Where I go for long documents, careful analysis and reasoning that has to hold together over length. Strong on Excel work and programming too. It can drift on very complex tasks, and some capabilities depend on your plan.
• 3. Microsoft Copilot — only if you live in Microsoft 365. This is a conditional buy. The integration across Excel, Word, PowerPoint, Outlook and Teams is the entire value proposition. If your organisation runs on Microsoft, it's excellent for administrative work. If it doesn't, skip it — for open-ended reasoning it's weaker than the two above, and the enterprise pricing is real.
• 4. Power BI with Copilot — for dashboards and KPIs. The right tool for management reporting and business analysis. One serious caveat: you still need to understand data modelling. If the underlying model is built wrong, the AI will confidently interpret it wrong, and a wrong dashboard is worse than no dashboard because people act on it.
• 5. NotebookLM — for studying documentation. The one people skip and shouldn't. It works on your documents: standards, manuals, technical documentation, regulations. For summarising and interrogating sources you have to know properly, nothing else in this list comes close. It's not a general business AI, and its output is only as good as the documents you feed it.
• 6. One specialist for your engineering discipline. This is where the list stops being generic — see below.

The engineering specialists

Generic AI listicles never include these, because the people writing them don't do engineering work. If your job is technical, this is where the real leverage is.

MATLAB Copilot and Simulink Copilot

Worth being precise here, because these are commonly conflated: they're two separate products. MATLAB Copilot works inside the MATLAB desktop environment; Simulink Copilot is a distinct product for Model-Based Design, introduced in Release 2026a. MathWorks also ships Polyspace Copilot for embedded code analysis.

For calculation, simulation, control systems and engineering models, this is powerful in a way general assistants aren't — because it works inside the environment rather than producing code you paste into it. The trade-offs are the obvious ones: a learning curve, real cost, and a scope narrowed to scientific and technical engineering. Best fit: electrical engineering, control, automation, simulation.

Autodesk Assistant, in Fusion

Again, the correct name matters: it's Autodesk Assistant, and as of writing it's available in Fusion as a Tech Preview — worth knowing before you build a process around it. Autodesk has it in preview across Fusion, AutoCAD, Inventor, Civil 3D, Revit and others.

Having AI inside the CAD environment for 3D design, manufacturing and prototyping is genuinely useful for task automation. It does not replace the judgement of an engineer or designer, and it's tied to the Autodesk ecosystem. Best fit: mechanical engineering, CAD and manufacturing.

The two workflows that actually matter

Owning the tools isn't the point — chaining them is. These are the two paths most of my work travels, and they're the most useful thing I can hand you:

For management reporting:

Excel → Copilot or Claude → Power BI → ChatGPT → management report

For technical work:

Data → Excel or Python → ChatGPT or Claude → MATLAB or CAD → technical report

Read left to right, each tool does the part it's actually good at and hands off. The general assistant sits in the middle as the connective tissue — interpreting, restructuring, drafting — while the specialist tools do the work at each end. The mistake is trying to make one tool carry the whole chain.

Which tool for which job

The rest, briefly

Tools I use or have used that didn't make the core six, with the honest one-line version:

• Perplexity — fast research with sources, good for comparing current information. It can synthesise sources poorly and doesn't always verify its own conclusions, so treat it as a starting point rather than an answer.
• Julius AI — genuinely simple for analysing a dataset and asking questions of it. Less complete than the big assistants, and it wants your data reasonably well structured.
• Powerdrill — useful for working repeatedly with business data because it holds context across datasets. Less established, smaller ecosystem, and worth reviewing its privacy terms and costs properly before committing company data.
• ClickUp AI — project management, tasks, documents and automation in one place. Can feel like too much: for a small team it's often overkill.
• Asana AI — cleaner and better organised for projects, owners, dates and workflows. Less flexible, and the advanced pieces sit behind higher tiers.
• Monday.com AI — very visual, strong on workflows and dashboards. Also gets complex, also costs, and some of its AI features are unnecessary at small scale.
• GitHub Copilot — covered in more depth in our coding assistants comparison.

The one-line version

If you want the whole thing compressed, this is how I'd describe each of these to a colleague in a corridor. Every one of them carries the strength and the catch together, because that's how they actually behave:

• ChatGPT — the Swiss army knife. It can also rust: versatile everywhere, the sharpest tool nowhere.
• Claude — a specialist at thinking, writing and coding, as long as you're willing to spend your time working around its ceiling.
• Gemini — the Google ecosystem plus AI. Ferrari speed, Mustang engine: quick and powerful, not precision-engineered.
• Perplexity — a researcher. One who hands you sources you still have to check.
• NotebookLM — studies your own documents. You have to be organised in order to be organised.
• Cursor and coding agents — programming, and actually enjoying your own code again.
• Midjourney — a very good graphics engine, until the wrong context gets in the way and spoils the run.
• Stable Diffusion and the open ecosystem — control and customisation, bought with complexity.

Notice that not one of those is unqualified praise. That's not me being difficult — it's the actual shape of this category. Every one of these tools is excellent at something and irritating somewhere else, and the whole skill is knowing which is which before you've paid for a year.

The caveat I'd insist on

Everything above has a shelf life. Opinions on these tools change fast, because the models and features change constantly — a limitation I describe today may be fixed in a release next month, and a strength may become a paid add-on.

So before you pay for anything here, check the current position yourself: price, usage limits, privacy terms, enterprise capabilities, and real user reports. Not the marketing page. That applies to this article too — check the date at the top before you trust the details.

Frequently asked questions

Do I really need more than ChatGPT?

For general work, often not — it's the most versatile single purchase. You need more when a specific job needs a specialist: dashboards, simulation, CAD, or working through your own technical documentation.

Is Microsoft Copilot worth it?

Only if your organisation genuinely runs on Microsoft 365. The integration across Excel, Word, Outlook and Teams is the whole value. Outside that ecosystem it's weaker than ChatGPT or Claude for open-ended reasoning and harder to justify on cost.

Can AI build my dashboards for me?

Partly, but you still need to understand data modelling. If the model underneath is built wrong, the AI will interpret it wrong with complete confidence — and people act on dashboards, which makes that failure expensive.

How many AI tools do I actually need?

Most people who use AI intensively settle on two to four, switching by task. I run at least five for engineering work and four for growth and design, because those chains have that many distinct steps. The number matters less than the rule: one versatile primary, then one specialist per genuinely different job.

What's the most overlooked tool here?

NotebookLM. If your work involves standards, manuals or technical documentation you need to actually know, working directly on your own sources is worth more than another general assistant.

Conclusion

The shift for me wasn't discovering a better tool than Word and Excel. It was accepting that one person is now expected to cover engineering, finance, operations and documentation, and that no single product does all of it well.

What works is a small stack with clear handoffs: a versatile primary assistant, a second for documents and reasoning, a specialist for whichever end of your work is technical, and something that reads your own documentation. Four to six tools, chained in the order above — not ten subscriptions bought one demo at a time.