Articles by Pedro Cunha
How your company gets recommended by ChatGPT and Google's AI answers
ChatGPT and Google's AI answers recommend the company that has the answer written on its own site and confirmed by third parties, in reviews, directories and lists. You can't pay your way into the answer, and an llms.txt file on its own changes nothing. On our site, appearances in Google's AI answers went from 262 to 1,482 in eight weeks and a manufacturer 900 km away reached us through ChatGPT, but contact through the site did not grow at the same pace.
19 min read
19 min read
Prompt injection: what an outside text can make your company's AI do
Any text an AI reads, whether an email, a WhatsApp message, a PDF or a transcribed voice note, can carry an instruction, and that is the first risk on the OWASP list for applications built on language models (LLM01:2025). No prompt can shield the model, because the rule and the attacker's text arrive through the same channel. The defense is to decide what the AI does on its own, what needs approval and what it never does, with the lock in code: at Balcão, the WhatsApp agent reads everything, but it has no way to start a conversation or do anything beyond replying, recording and handing off to a person.
17 min read
17 min read
AI agent on WhatsApp Business: what you pay for, what is free and what changes between the official and an unofficial API
Choosing between the official WhatsApp API and an unofficial library such as Baileys is not only about price: on the official API, Meta's rules shape the service, with a 24-hour window, approved message templates, one integration per number and per-message billing; on an unofficial one those rules disappear and what remains is the risk of the number being banned, because the use breaks WhatsApp's terms. The official API is not just the expensive option, since every number gets 1,000 free replies a month, and the unofficial one is not free. On the WhatsApp customer service platform we run, we operate both and choose client by client.
18 min read
18 min read
Why your team stops using the new system and goes back to spreadsheets
A team goes back to the spreadsheet, the notebook or WhatsApp when the new system takes more work than the old way on the task it does every day, not because people resist change. Adoption is decided on the screen of the most repeated action, in what each role sees and in a rollout done alongside the team, and it is measured after launch by the work done in the system, not by logins. At Fitness Academia, the client created all 15 user accounts on its own and entered its targets through December inside the tool.
15 min read
15 min read
Code review with AI: what it catches and what slips through
AI code review is good at logic errors, side effects in other parts of the system and obvious security flaws. It does not catch business rules nobody wrote down, architecture decisions or what only the team knows about production. Across 26 pull requests in our projects, between August and September 2026, it flagged 51 problems before merge, and the largest group was wrong output on screens and reports. It only works when it reviews against the team's written standard, and the merge is always approved by a person.
15 min read
15 min read
The notification map: the document that settles every system alert before the code
A notification map, also called a notification matrix, is the document that lists, before development, every alert a system sends: the event that triggers it, who receives it, through which channel, at what moment, with what text, what happens if it does not arrive and who must not receive it. In the health app we are modeling, it holds 78 notifications, and users can turn off only 3 of them. Written before the code, it prevents alerts to people on leave, verification codes over WhatsApp stuck waiting for Meta approval and sensitive data on a locked phone screen.
15 min read
15 min read
Why TestFlight is not a way to distribute your company's app
TestFlight is Apple's testing tool: each build lasts 90 days and then becomes unavailable to whoever installed it, so it cannot be the way your team uses an app every day. The right channel depends on who uses the app. A public app goes to the store; your own team's app goes to the store behind a login the company provides, or ships as a private or unlisted app; an app built for a client company ships as a private app for that organization, on Apple and on Google. Apple's Enterprise Program is not a shortcut: it requires 100 or more employees and is for internal use only.
13 min read
13 min read
How to keep an anonymous whistleblowing channel truly anonymous
An anonymous reporting channel loses its anonymity in places the form never shows: the IP address recorded by the hosting platform, the location inside a photo, the author inside a PDF, third-party analytics and bot checks, and the combination of department, shift and date in a small team. Each of these has a technical fix that has to be in the system from the scope onward. In the reporting channel we are building for a client, the portal was moved off Vercel because of the IP, every attachment loses its metadata before it is stored, and follow-up with the reporter happens through a case code and password, with no email.
16 min read
16 min read
Already paying for off-the-shelf software: configure, integrate or build your own
If you already pay for off-the-shelf software and something is missing, you have three ways out, in this order of cost: switch on what the software already offers, integrate around it or build a module alongside it, and build your own system. In one audit we ran, three of a client's four requests already existed in the software they were paying for, switched off. Your own system pays off when the process is what your customers buy from you, and even then it can live alongside the off-the-shelf software instead of replacing it.
15 min read
15 min read
When your app pays a commission to Apple and Google, and when it can charge outside the store
An app pays the store a commission when it sells something digital that people use inside it, such as a subscription, content or an unlocked feature; physical services, live consultations between two people and free apps that accompany a paid web system charge outside the store, with no commission. In Brazil, since June 2026, Apple charges 21% plus a 5% processing fee on purchases through its system, and still takes a commission when digital content is paid for outside it. The most expensive mistake is mixing, in the same app, what is exempt with what is not.
16 min read
16 min read
How much it costs to put AI to work inside a company system, and what makes the bill go up or down
When AI runs inside a company system, you pay per token, and a customer service conversation costs cents: on Balcão, our WhatsApp customer service platform, triage comes to about R$ 0.15 (roughly US$ 0.03) per conversation. What weighs most is the text the AI rereads on every reply, and prompt caching bills that part at 10% of the price or less; model size comes next, with up to a 100x gap between the largest and smallest model from the same company. The bill also changes when you measure the result: in the simulation of a sales agent, the model that cost four times more brought more leads to the call, and the math by result came out in its favor.
17 min read
17 min read
When an app built on Lovable needs to leave Lovable, and when it doesn't
Leaving Lovable can mean moving the hosting, moving the database out of Lovable Cloud or no longer building with the tool, and the first two work without the third: at Sistema Forja, infrastructure went from about R$ 2,000 to about US$ 20 a month and the client kept editing the app in Lovable. If you are still validating the idea, stay where you are. Rebuilding only makes sense for one part, when search is the product or when money flows through the app.
15 min read
15 min read
Mobile app or web system: how to decide what goes on the phone
The question is answered per user profile, not for the whole system. An app makes sense for people who work away from a desk and need what only an app does well: keep working with no signal, talk to a device over Bluetooth, use location with the screen off. Whoever manages the system, uses it now and then or arrives through a link stays on the web. In our projects that have an app, the web system keeps doing everything and the app handles only the field work: at Beckers, the technician fills in the report in an Android app and the office reads everything in the web system.
16 min read
16 min read
How to organize company knowledge so AI can work with it
For AI to work the way your company works, you do not need to train a model. You need to organize what the company knows in a place the AI can read while it works. We use Tiago Forte's second brain method: capture everything that comes in, organize it into four folders (projects, areas, resources and archives), write each process down only once and split the work with a simple rule. AI takes the volume, which is capturing and organizing, and people keep the judgment, which is deciding and producing.
17 min read
17 min read
AI chatbot, assistant or agent: how to pick the right level for each task in your company
Chat, assistant and agent are three levels of delegation to AI. In a chat, you ask and it forgets. An assistant has learned one of your processes and helps when you call it. An agent decides on its own when to act and where to leave the result. What picks the level is how much you already trust what it delivers and what a mistake costs, not how sophisticated the task is: the most important process in our company, the project quote, runs entirely at the assistant level, and the contract that used to take up to 40 minutes to put together now takes 5.
18 min read
18 min read
Spreadsheet, off-the-shelf ERP or custom software: the answer is almost always more than one
A spreadsheet, an off-the-shelf ERP and custom software are not competing for the same spot: they sit in different layers of the same operation, and almost every healthy company runs all three at once. The expensive mistake is not picking the wrong one, it is changing layers when you did not have to: at a five-location gym chain the off-the-shelf ERP stayed as the source of every sale, and the custom system simply started reading from it every five minutes. The spreadsheet, in turn, stops working for three objective reasons, and none of them is size.
14 min read
14 min read
What decides the platform for a company website
The platform decision for a company website in 2026 is no longer builder versus custom: it is AI generator versus professional development. We no longer recommend a site builder for any new project, and for a small site with no personal data and no search as a sales channel, an AI generator is enough. The case leaves the generator the day someone becomes accountable for performance, SEO, GEO, AEO, security and data protection.
12 min read
12 min read
How to make AI follow your company's method
Constraining a model to your company's methodology is not writing a better prompt or training your own model — it is taking out of its hands whatever cannot go wrong. Four layers: the method written outside the prompt and read at run time, the arithmetic in code, the banned list along with the vocabulary that replaces it, and a brake that stops the action instead of asking for good behaviour. In the Sistema Cuidar report that means AI writes 5 of the 15 sections and not a single number in the document came from a language model.
19 min read
19 min read
Who owns the App Store account: you or the company building your app
Owning the app is three things decided in three different places: the code, in the contract; the store listing, in the developer account; and the ability to ship the next version, in the signing certificates — and whoever controls the account controls the product. What matters isn't which account the app is born in, but which account it ends up in, whether yours started being opened on day one, and whether the migration is a planned step in the contract. We've run that full cycle: the transfer preserves ratings and your installed base, and it breaks specific things that need to be mapped beforehand.
16 min read
16 min read
Integrating with an ERP: what the other system decides for you
The architecture of an integration is not your choice: it is imposed by what the other system offers. Across three integrations we built — NextFit, Altimus and Controlle — the same question (what changed over there?) had three different answers, because the sources were different. And whatever the source does not provide becomes a limit on your product, not just a line of code.
16 min read
16 min read
What can be salvaged from AI-generated code
What breaks in an AI-built product isn't ugly code: it's the boundary. In the app we audited, all eight server endpoints accepted unauthenticated requests — payment endpoints included — and any user's private data was readable by everyone else. None of it required a rewrite: the client kept shipping with the same AI after the foundation was fixed.
12 min read
12 min read
When a legacy system needs replacing — and when it doesn't
Old, ugly or slow is not a reason to replace: of the nine systems that were already running when we arrived, we replaced two and still maintain five. The signs that actually decide are different ones — nobody can change the system, you lost access to your own product, the team has already built a path around it. And before any of them, an audit of what you already have: in a recent case, three of the client's four requests already existed in the current system, switched off.
26 min read
26 min read
What changes in a system when the app has to work with no signal
Offline-first is not an app feature: it is a decision that changes the database, the server and the product scope — above all, what does not go onto the device. And an app that promises offline and freezes is worse than an honestly online one: the difference between the two lives in details that almost never show up in a proposal. Almost every write the server later rejects is a rule the app did not know, and the first fix is to move that rule into the app.
19 min read
19 min read