AI or a developer? Often both.
AI tools now do a lot on their own. Here is an honest map: what they do well without a developer, where they stop, and what changes when someone builds around them. If AI alone is enough for you, I will tell you.
Pick a level. See what it does, and where it stops.
ChatGPT, Claude, Gemini, Copilot
Great for
- Drafting emails, posts and documents
- Summarising, translating or analysing a file
- Brainstorming and first ideas
- A one-off formula or a small script
Where it stops
- It cannot see or change your own systems
- You copy and paste in and out, every time
- Your data goes to an outside service
- Nothing happens unless someone asks
Example“Summarise this supplier contract and list the risky clauses.”
Zapier, Make, chatbot builders, AI features in your software
Great for
- Connecting popular apps to each other
- Simple automations: form → summary → email
- A basic chatbot that answers your FAQ
Where it stops
- Your own or older software is out of reach
- Rules with many exceptions turn into a maze
- Every task run is billed: costs grow with volume
- Hard to debug once there are dozens of flows
Example“When a quote request arrives, summarise it and post it in the team channel.”
Custom software, with AI where it pays off
Great for
- AI wired into your real data and software
- Actions with safeguards: approvals, risk checks
- AI running on your own machines when data is sensitive
- Monitoring, tests and maintenance: it keeps working
- The right model for each task, so costs stay low
- Built around the way your team actually works
Where it stops
- It is a project: weeks, not minutes
- Worth it when the problem repeats or matters
Example“Answer customers from our live catalogue and stock, add to the cart, and hand over to us when needed.”
| AI on its own | No-code + AI | A developer + AI | |
|---|---|---|---|
| Ready in minutes | |||
| Uses your real data and software | |||
| Takes actions, with safeguards | |||
| Your data stays in-house | |||
| Copes with volume, day after day | |||
| Costs stay predictable as you grow | |||
| Fits the way your team works | |||
| Someone accountable when it breaks |
Two dots: yes · One dot: partly · None: no
Cheap to start. Then the bill grows with you.
Without a developer, every new need adds a subscription, every extra task adds to a usage bill, and nobody keeps the AI from using the most expensive model for everything. Follow each line to see where the costs creep in.
Hover or tap the points · Illustrative: shapes, not prices.
How I keep the bill flat
- The right model for each task: a small, fast one for routine questions, a large one only when it is needed.
- Answers cached and reused instead of paid for twice, as in Linguise.
- Limits and monitoring, so a loop or a spike never runs up a bill overnight.
- One tool that does the job instead of five subscriptions stitched together.
Not theory. Things I have built.
Each of these is the line between “AI that talks” and “AI that does the job”, in a project you can look at.
Liliwi answers shoppers from the store’s live catalogue and stock, not from general knowledge.
See Liliwi.ai SafeguardsAphelia proposes each server command with a risk score and waits for a human to approve it.
See Aphelia PrivacyThe anonymisation software runs its AI on the computer itself. Nothing is sent outside.
See AnonyShield Volume and costLinguise serves millions of translated pages a day: bots filtered, pages cached, AI used only for new content.
See LinguiseSometimes the right answer is a subscription.
- The task happens once, or once in a while.
- It is mostly writing, reading or summarising.
- Your apps are popular ones that a no-code tool already connects.
- Nobody loses money or time if it fails now and then.
If that is your case, I will say so on the first call and point you to the right tool. You will have lost twenty minutes, not a budget.
AI writes the routine. I take care of the rest.
AI drafts much of the everyday code, tests and documentation. That frees my time for what decides whether a project works: understanding your problem, designing a tool your team enjoys using, choosing the architecture, and reviewing every line before it reaches you.
You get software faster than a few years ago, with the same person accountable for it.