AI will write your software. But who will run it?

AI has made the most expensive part of IT cheap: writing software. Internal tools, customer portals and automations that once needed a team of developers can now be built by a small group in a fraction of the time. Smaller and mid-sized companies will therefore have their own IT departments that write their own software.

One thing hasn’t got cheaper, though. Every new application needs servers or cloud, deployments, backups, monitoring and security. And the people who can do that well are getting scarcer. This article shows why operations are becoming the bottleneck – and what companies that can’t afford their own DevOps team can do about it.

AI writes code. And it will write more and more of it

Gartner expects that by 2028, 90% of enterprise software engineers will use AI coding assistants, up from less than 14% in early 2024 (Gartner).

Stack Overflow’s 2025 survey of nearly 50,000 developers shows the shift is already underway:

84% of developers use or plan to use AI tools, up from 76% a year earlier (Stack Overflow).
51% of professional developers use AI every day (Stack Overflow Survey 2025).
But 46% of developers don’t trust AI’s accuracy, up from 31%. The top complaint: solutions that are “almost right” and take longer to debug.

And it’s no longer only developers who write software. Gartner estimates that in 2026, people outside formal IT departments will make up at least 80% of users of low-code development tools, up from 60% in 2021 (FutureCIO citing Gartner).

For smaller companies this means one thing: custom applications are no longer a luxury of the big players. Routine work done by hand today will be taken over by tools the company writes itself.

Every application grows what keeps it alive

Code is only the beginning. An application needs somewhere to run, a database, backups, deployments of new versions, monitoring and protection. The more applications a company has, the more of all this there is – and the more it costs.

Cloud spending shows it. Gartner put worldwide public cloud spending in 2025 at $723.4 billion, 21.5% more than the year before (Gartner). In 2026 the market is expected to grow another 21.3% and reach $1.48 trillion by 2029, driven mainly by AI (Gartner).

Waste grows too. In Flexera’s 2026 report, companies estimate that 29% of their cloud spend is wasted – the first increase in five years. 85% of respondents named managing cloud spend as their main challenge (IT Brief on Flexera 2026).

And with AI comes risk. Veracode tested over 100 language models on 80 coding tasks: in 45% of cases the generated code contained an OWASP Top 10 vulnerability, and for Java it was over 70% (Veracode). Software written faster also means more places that need watching.

Nobody to run it

IDC predicts that by the end of 2026, over 90% of companies worldwide will feel the IT skills shortage, with losses from delayed projects and lost business of up to $5.5 trillion (IDC). Right after AI skills, IT operations skills are the most in demand (IT-Online on the IDC study).

Czechia is worse off than the EU average. According to Eurostat, 70.5% of Czech companies that looked for an ICT specialist in 2023 had trouble filling the role. The EU average was 57.5%, and only Germany was higher (Eurostat).

And an in-house team has its price. According to the Czech Statistical Office, the average gross salary of an ICT specialist in the private sector was CZK 102,000 a month in 2025 (ČSÚ). Experienced DevOps engineers usually earn more than average. A rough estimate:

Option	People	Employer cost per year
One person for everything	1	approx. CZK 1.6 million
Small team with cover	3	approx. CZK 4.9 million

The estimate uses the average ICT specialist salary and 33.8% employer contributions, excluding hiring, training, tools and on-call pay. For a company with a few applications, that’s a lot of money for work that won’t keep a whole team busy.

One person isn’t enough

A typical smaller company has one admin for its infrastructure, two at best. It works until that person wants a holiday, gets sick, leaves for a competitor, or an incident hits at 3 a.m. Then it turns out only they know the systems and nothing is documented.

Pressure from outside is growing at the same time:

Attacks target the smaller ones. According to Verizon’s DBIR 2025, ransomware was part of 44% of all data breaches. At small and mid-sized businesses it was 88%, at large ones 39% (The Record on Verizon DBIR 2025).
Attackers exploit unpatched systems. Exploitation of vulnerabilities as an entry point rose by 34%, and breaches through third parties doubled to 30% (Verizon).
Security is now a legal obligation. Since November 1, 2025, Czechia’s new Cybersecurity Act No. 264/2025, implementing the NIS2 directive, has been in force. It covers more than 6,000 organizations, up from roughly 400 (Lupa.cz, TZB-info). Management is accountable, and in the higher regime incidents must be reported within 24 hours (Forvis Mazars).

Security is no longer a project you do once. It’s ongoing work: tracking vulnerabilities, patching, testing backups, watching access and keeping audit evidence. One person can’t do that on top of day-to-day operations.

What are the options

A company that writes its own software has essentially four ways to run it. None is universally right – they differ in cost, in how well they handle the loss of a key person, and in how much attention they need from management.

Option	Strength	Weak spot	When it makes sense
In-house team	Full control and company knowledge	Hiring, cost, cover only from 3 people	Large infrastructure that keeps a team busy
One admin or freelancer	Low cost	Single point of failure, security and audit beyond one person	Very small, simple environments
External operations team	A full team with cover, pay for what you need	Vendor dependency if you don’t own code and access	Small and mid-sized companies with their own development
Hosted on a provider’s platform	Least hassle, the provider handles everything under the app	Less freedom in technology	Companies that don’t want to deal with infrastructure at all

Large companies are already going this way. According to Flexera’s 2025 report, 60% of organizations are turning to managed service providers to get cloud spend under control (BetaNews on Flexera 2025).

With external operations, one thing decides: that it doesn’t create a new dependency. Repositories, cloud accounts, access and documentation should belong to the company. Then the provider can be replaced, or operations taken back, at any time.

What to take away

The coming years will bring more software, more infrastructure and stricter rules. By 2028, 90% of developers are expected to code with AI, the cloud market will more than double between 2025 and 2029, and NIS2 already affects thousands of Czech companies. But the number of people who can run all this won’t grow at the same pace.

For smaller and mid-sized companies, that means five practical steps:

Map what you run. Which applications, where they run, who has access, and who can restore them.
Calculate the real cost of operations. Including cover, on-call, and the time developers spend on operations today.
Decide what to keep and what to hand over. Development can stay in-house; operations can be run by someone who does it every day.
Insist on a standard. Infrastructure as code, separate environments, monitoring, tested recovery, documentation – whoever runs operations.
Check NIS2. The Czech cybersecurity agency NÚKIB offers a calculator on its portal that tells you whether the act applies to you. For a quick check, try our NIS2 guide.

With AI you can write software faster than ever. But to run reliably, securely and even while your admin is on holiday, it needs a solid foundation. That foundation is what’s scarcest today.