Guide: AI in business › AI implementation, tools and team skills
An AI software house — a software development company specialising in artificial intelligence — will build a system to specification. An AI agency starts earlier, with the question of which decisions and processes in the company need improving, and is responsible for the outcome in the process, not for the code. The choice depends on whether you already know exactly what needs to be built. Below you’ll find a comparison, costs, checklist questions and situations in which it’s best to combine both approaches.
In 60 seconds Choosing a partner at a glance
- The difference in one sentence: a software house writes a program, an AI agency changes the way you work and measures the effect, and AI consultancy ends with a plan.
- When to choose a software house: you have a documented process and requirements, and you need your own system or the integration of multiple systems.
- When to choose an AI agency: you know where you’re losing time or money, but you don’t know what tool will solve it.
- Budget: the cheapest way to start is with a diagnosis and a pilot on off-the-shelf tools; building your own system is a separate, bigger expense.
- The biggest risk: a great system nobody uses, and no single party responsible for the overall outcome.
What is an AI software house, and what is an AI agency?
AI software house is a software development company that builds solutions based on artificial intelligence: applications, integrations, custom models, deployment of large language models (LLMs, i.e. models like those in ChatGPT or Gemini) on the company’s own infrastructure. It works to specification and is responsible for making sure the program works.
An AI agency starts with the business problem. It asks where the company is losing time or making decisions by gut feeling, selects off-the-shelf tools, sets working rules and measures the effect. There’s also a third route, AI advisory: consultancy that ends with an AI strategy and a plan, but doesn’t include the implementation itself.
In Poland, this choice still concerns a minority of companies. According to Eurostat, 8.4% of Polish companies with at least 10 employees used AI in 2025, against an EU average of 20.0%, so most companies are only just facing their first implementation and the question of who to entrust it to. If you’re only just getting to grips with the whole subject, start with our guide to AI in business, and come back here when it’s time to choose a partner.
Na rynku obie nazwy często się mieszają, a firma wdrażająca AI może mieć w nazwie „agencja”, choć w praktyce głównie programuje. Dlatego patrz na to, od czego zaczyna rozmowę, a nie na szyld.
AI software house, AI agency and consultancy compared
The three types of provider are easiest to compare by what they start with and what they take responsibility for.
| AI software house | AI agency | AI consulting | |
|---|---|---|---|
| Starts with | the specification | the business problem | strategy and organisation |
| Delivers | a system, integration, code | implementation in the process, working rules, measurement | recommendations, a plan, tool selection |
| Responsible for | the software working | the outcome in the process | the quality of the analysis |
| Best when | you know exactly what to build | you know what hurts but not how to fix it | the board is setting direction for years ahead |
| Risk | a great system that nobody uses | too little technical depth for a complex build | a plan without execution |
When should you choose an AI software house?
A software house is a good choice when the problem is already well defined and the solution requires a lot of programming.
- You have a documented process and requirements, and you need your own system or an integration.
- The project requires a lot of programming: an application, a custom model, connecting multiple systems, AI working on data that must not leave the company.
- You’re planning an AI implementation on your website that requires integration with company systems, e.g. an assistant that checks order status in the ERP system.
- You have someone in the company who can translate business needs into a specification and sign off the work.
When should you choose an AI agency?
An AI agency works well where there are more people and processes than code. Most often this concerns marketing, communication, sales or reporting.
Tak zaczynamy w Commwork®AI: pierwszy miesiąc to mapa procesów i programów, które firma już ma, i tego, ile z nich naprawdę używa. Potem zarząd dostaje kilka wariantów rozwoju z wyliczeniem kosztu i efektu, a dodatkowe narzędzie doradzamy tylko przy bardzo zaawansowanych zadaniach.
- You know where you’re losing time or money, but you don’t know what tool will solve it.
- You want to test off-the-shelf tools quickly before commissioning a build, because many gen AI solutions for business work without a single line of custom code (you’ll find an overview in our round-up of AI tools for business by task).
- The implementation is primarily about people and their habits, and only secondarily about code.
- You need to measure the effect and set rules for how the team works with AI, for example an AI policy and training.
- You’re looking for a partner to roll out Copilot or ChatGPT in your team. A good Copilot implementation partner doesn’t stop at configuring licences, but teaches people how to work with the tool and measures the effect — we write about this in more detail in the context of rolling out Copilot in your company.
Jak Commwork®AI pomaga wybrać między budową a zmianą procesu
Najdroższy błąd przy wyborze wykonawcy to zamówienie budowy, zanim ktoś sprawdzi, co da się zrobić programami, które firma już ma. Commwork®AI łączy te programy w jeden system, a zespół Commplace® mapuje procesy i przygotowuje warianty z kosztem i efektem. Wariant wybiera zarząd.
- Diagnoza i pilotaż 18%
- Licences for off-the-shelf tools 22%
- Build and integrations 35%
- Training and working rules 10%
- Utrzymanie i rozwój 15%
- Week 3Mapa 18 procesów i programów firmy gotowa
- Tydzień 5Pomiar czasu odpowiedzi na zapytania: 3 dni
- Tydzień 6Test gotowych narzędzi na prawdziwych zapytaniach
- Week 8Trzy warianty rozwoju z kosztem i efektem dla zarządu
- Week 12Decyzja: gotowe narzędzie czy budowa brakującej części
- 2.12.2027AI Act: obowiązki dla systemów wysokiego ryzyka
- 01Mapa procesów
- 02Warianty z wyliczeniem
- 03Decyzja zarządu
- 04Pilotaż i połączenia
- 05Measuring the impact
- W pierwszym miesiącu mapuje procesy i programy, które firma już ma, i pokazuje, ile z nich realnie wykorzystuje.
- Łączy te programy w jeden system i pokazuje wynik w kokpicie, w którym każdy wskaźnik ma źródło, datę, próg i właściciela.
- Prowadzi rejestr decyzji: kto, kiedy, na jakich danych i z jakim skutkiem, a miejsca, w których decyduje człowiek, ustala się tak jak przy implementing AI in marketing and PR.
The view reproduced is modeled after the dashboards provided to customers. The names, numbers, and terms are examples and show the way the system works, not the results of a specific company.
The hybrid model: an agency and a software house working together
Często najlepiej działa współpraca obu firm naraz, z jasnym podziałem ról. Agencja diagnozuje, wybiera kierunek, prowadzi pilotaż na gotowych narzędziach i pisze wymagania, a software house buduje to, czego gotowe narzędzia nie zapewnią. Jeden warunek jest nie do negocjacji. Za całość efektu odpowiada jedna strona, bo inaczej każda będzie wskazywać na drugą. Nowe narzędzie to też zmiana sposobu pracy ludzi, więc przyda się plan, o którym piszemy w tekście o tym, jak skutecznie zarządzać zmianą w organizacji.
The second condition is human oversight, known as human in the loop. Whoever does the build, agree in advance at which points AI only prepares and a person decides: publication, spending, commitments to customers. At Commplace we call this a human in the decision loop. Be especially careful about this when you commission an AI agent for your business, i.e. a program that performs several steps in sequence on its own; we show what such an agent can do and how it’s built in the guide how to create an AI agent.
How to choose an AI implementation partner, step by step
The order matters. Problem and measurement first, then the tool, and finally a build if needed. We set out a general plan for the first three months, from diagnosis to the board’s decision, in our article on how to implement AI in your company in 90 days.
- STEP 01Name the problem and the metric
Write down in one sentence what needs to improve and how you’ll measure it, for example “respond to a request for quotation in 1 day instead of 3”.
- STEP 02Check off-the-shelf tools
Before you commission a build, test available solutions for 2–4 weeks on real tasks. AI consultations for businesses at this stage often save an entire development project, because it turns out an off-the-shelf tool covers most of the needs.
- STEP 03Decide what’s left to build
If off-the-shelf tools cover most of your needs, you only build the missing part. If not, you already have requirements for a software house, written on the basis of real work rather than assumptions.
- STEP 04Compare providers on the same questions
Send the same questions to two or three companies (the list is further down in this article) and compare the specific answers, not the presentations.
- STEP 05Agree a pilot with a success criterion
The first contract covers a small, closed scope with a success metric and an evaluation date. Only then a bigger budget.
How much does AI implementation by an agency or a software house cost?
There’s no single rate, because these companies sell different things, so it’s sensible to start with a diagnosis and a short pilot and only decide on a bigger budget after that. For an idea of tool costs: on Microsoft’s Polish website, Microsoft 365 Copilot for business costs from EUR 15.60 net per user per month with annual billing (promotional price until 31 December 2026, as of 24 September 2026).
When many variables are uncertain, a range with probabilities is better than a single figure, calculated for example using the Monte Carlo method, i.e. by simulating many cost scenarios.
Koszt wdrożenia AI rośnie najszybciej nie przy licencjach, tylko wtedy, gdy budowa zaczyna się przed pilotażem. Chcesz najpierw wiedzieć, czy w ogóle potrzebujesz programowania? Agencja AI Commplace zaczyna od diagnozy procesów i daje wymagania, które możesz potem zanieść do dowolnego software house’u.
| Item | AI agency | AI software house | What it depends on |
|---|---|---|---|
| AI diagnosis and strategy | usually the first stage | rarely offered | number of processes and departments |
| Licences for off-the-shelf tools | selection and configuration | usually out of scope | number of users and plan |
| Build and integrations | limited, often with a partner | the main part of the budget | number of systems, data security |
| Model usage fees | included in the subscription for off-the-shelf tools | separately, based on usage | number of queries, chosen model |
| Training and working rules | in scope | usually out of scope | team size |
| Maintenance | updating rules, measurement | fixes, development, servers | pace of change at model providers |
10 questions to ask before choosing a provider
Te pytania działają niezależnie od tego, czy rozmawiasz z agencją, czy z software house’em. Liczy się nie tyle treść odpowiedzi, ile to, czy jest konkretna: dobra odpowiedź o pilotażu zawiera miarę, termin oceny i nazwisko osoby, która oceni wynik, a samo „to zależy” bez dalszego ciągu jest sygnałem, żeby dopytać. Część z nich zadasz każdemu partnerowi zewnętrznemu, więc przydadzą się też kryteria, które opisaliśmy przy temacie how to choose a marketing agency.
- What problem are we solving, and how will we measure that it has been solved?
- Can it be done with an off-the-shelf tool before we start building?
- Who on our side signs off the work and makes decisions?
- What data will be needed, and where will it be processed?
- What does AI do on its own, and what requires human approval?
- What does the pilot look like, and how will we know it has succeeded?
- Who maintains the solution after implementation, and how much does it cost?
- What do we own: code, data, instructions?
- How does the solution meet the requirements of the AI Act?
- What similar implementations can the provider show, with results rather than just a description?
When the software house is looking for an agency: AI visibility for IT companies
There’s also the reverse situation. A software house looks for help itself, because customers begin choosing a provider by asking ChatGPT, and the model recommends the competition. The query “which SEO agency for AI for a software house” is appearing in search more and more often.
The same principle applies here as in other industries. A Muck Rack study from May 2026 showed that 84% of sources cited by models (ChatGPT, Claude, Gemini) are earned media, i.e. publications the company didn’t pay for. For an IT company, this means independent trade publications, case studies describing results, expert talks and solid profiles in provider directories. We set out what to look for when choosing an agency for this task in 15 questions to ask a GEO or AI agency; you can read about PR itself in this industry in the article PR activities in the IT industry.
Commplace works with technology companies as part of its specialisation technology and IT.
Wdrożenie sztucznej inteligencji w firmie w 90 dni na przykładzie
Przykład poglądowy, bez konkretnego klienta. Firma handlowa zatrudniająca 120 osób odpowiada na zapytanie ofertowe średnio w trzy dni i chce skrócić ten czas do jednego. W pierwszym miesiącu partner do wdrożenia AI w firmie rysuje mapę procesu: skąd przychodzą zapytania, kto je kwalifikuje, gdzie leżą cenniki i historia klienta. W drugim zespół testuje gotowe narzędzia, czyli asystenta w pakiecie biurowym i proste połączenie skrzynki z systemem klientów, a przy okazji mierzy czas odpowiedzi. W trzecim zarząd dostaje wynik pilotażu i decyzję do podjęcia: zostać przy gotowych narzędziach czy zlecić software house’owi budowę brakującej części, na przykład wyszukiwania w dokumentach firmy. Jak ten sam problem wygląda od strony handlowej, opisujemy w tekście o tym, jak mierzyć i poprawić sales efficiency.
Zanim podpiszesz umowę z którymkolwiek wykonawcą, sprawdź, czy masz te rzeczy. Jeśli brakuje dwóch lub więcej, zacznij od diagnozy, a nie od budowy.
Gdzie leży próg decyzji między budową a gotowym narzędziem? Jeśli po pilotażu gotowe rozwiązanie pokrywa większość potrzeb i miara się poprawiła, dobuduj tylko brakujący element. Gdy dane nie mogą opuścić firmy albo trzeba połączyć kilka systemów, a gotowe narzędzie nie ruszyło miary, pora na software house i na wdrożenie dużych modeli językowych LLM we własnym środowisku. Przy procesach, które mają się dziać bez udziału człowieka, przeczytaj najpierw, czym jest agentic AI and AI agents, a przy prostszych automatyzacjach sprawdź, co da się zrobić w automation with AI and n8n, zanim zamówisz kod.
- jedno zdanie z problemem i miarą, na przykład czas odpowiedzi na zapytanie,
- pomiar tej miary sprzed wdrożenia, żeby było z czym porównać,
- osoba po stronie firmy, która odbiera pracę i ma czas na pilotaż,
- lista systemów i danych, na których AI ma pracować, z informacją, co nie może wyjść poza firmę,
- zasady, co AI robi samo, a co zatwierdza człowiek,
- budżet na utrzymanie po wdrożeniu, nie tylko na budowę.
When to start, and when to hold off on choosing
Dobry moment na wybór wykonawcy to chwila, gdy zespół ma czas na pilotaż, a zarząd jest gotów wyznaczyć właściciela projektu. Złym momentem jest szczyt sezonu sprzedażowego, trwająca reorganizacja albo sytuacja, w której nikt nie wie, na jakich danych AI miałaby pracować. Jeśli chcesz przejść diagnozę i pilotaż z partnerem, który zaczyna od procesu, a nie od kodu, zobacz, jak pracuje Commplace AI agency.
Also check the regulatory calendar. Since 2 August 2026, the AI Act’s transparency rules have applied, for example the obligation to inform customers that they are talking to a chatbot, while obligations for high-risk systems, including in recruitment, have been postponed to 2 December 2027. A provider building an AI system for you should know these dates without looking at their notes; we’ve compiled the full calendar in our article on when the AI Act applies and who it affects.
The costliest mistakes when choosing a partner
Many needs are covered by tools available on subscription, so by commissioning a build blind, you pay for months of programming and a budget spent on something that could have been tested in two weeks.
A software house works to specification. If no one on the company’s side keeps an eye on it, the wrong thing gets built, followed by fixes, scope disputes and delays.
The agency is responsible for the process, the software house for the code, and nobody for the outcome. The cost: the system works, the business result doesn’t materialise, and each party points at the other.
Models and tools change every few months, so a solution without planned updates stops working or becomes more expensive after six months, and money for fixes then has to be found in emergency mode.
The demo always works. The cost: a big contract signed before anyone has tested the tool on the company’s real data.
What companies ask before signing an AI contract
What is an AI software house?
It’s a software development company that builds solutions based on artificial intelligence (applications, integrations, deployments of language models on company infrastructure), works to specification and is responsible for the program working, not for changing the process. Code is its product.
How much does AI implementation by an agency cost?
It depends on the scope. It’s sensible to start with a diagnosis and a short pilot on off-the-shelf tools, and only then decide on a bigger budget. Building custom integrations tends to be the most expensive part, and the team’s time is the most often underestimated. With us, an AI implementation diagnosis and strategy costs from PLN 8,000 net, and the first conversation is free of charge (as of 24 September 2026).
Can you combine a software house and an AI agency?
Yes, and it is often the best route: the agency diagnoses, chooses the direction, runs a pilot on off-the-shelf tools and writes the requirements, and the software house builds what off-the-shelf tools cannot provide. What matters is that one party is responsible for the overall outcome.
What is AI consulting?
It’s help in deciding whether to implement AI at all, and if so, where and how: process analysis, tool selection, risk and cost assessment, working rules and regulatory compliance. Good advisory ends with a plan with owners and deadlines, not just a presentation.
How do you choose a partner for implementing AI in your company?
Ask two or three providers to answer the same questions: what problem are we solving, how will we measure the effect, where will the data be processed and who will maintain the solution. Start with a small pilot with a success criterion, and only then sign a bigger contract.
Do you need a software house to roll out Copilot?
Usually not. Copilot works within Microsoft 365 tools, so the rollout is mainly about configuring permissions, working rules and training the team. A software house only becomes useful when Copilot needs to use data from other company systems.
Which agency can help a software house with AI visibility?
One that combines PR with content and measurement. AI models mainly cite external publications, so optimising your website alone isn’t enough. You need independent trade articles, case studies and regular measurement against a fixed list of questions your customers ask.
Sources
- SOURCEEurostat: 20% of EU enterprises use AI technologies · 11 December 2025
- SOURCEZFPR: survey on the use of AI in PR agencies · 29 June 2026
- SOURCEMuck Rack: What Is AI Reading? (May 2026) · 7 May 2026
- SOURCEMicrosoft: Microsoft 365 Copilot pricing for business (Poland) · as of 24 September 2026
- SOURCE
- SOURCEEUR-Lex: Regulation (EU) 2026/1744 (Digital Omnibus on AI), Polish text · of 8 July 2026, OJ EU of 24 July 2026, accessed 24 September 2026
Read next
Dostaniesz diagnozę: gdzie AI da Twojej firmie najwięcej i jak to zmierzymy. Jeśli potrzebna będzie budowa, otrzymasz gotowe wymagania dla software house’u. Pierwsza rozmowa jest bezpłatna.
Sebastian Kopiej, President of Commplace®. In public relations since 1996. Text prepared with the help of AI tools and fact-checked by the author. Data current as of September 29, 2026.
Read more on this topic