Why are companies building their own software again? What the Starbucks case signals.

What exactly is Starbucks doing?

On July 9, 2026, Bloomberg News reported that Starbucks is using AI to build in-house alternatives to software it currently buys. The story draws on an internal presentation and a recording of an employee forum.

Its chief technology officer, Anand Varadarajan, who joined from Amazon in January 2026, told that forum the company spends about $400 million a year on software alone and that “there’s clear opportunities to reduce the spend in software.” Starbucks declined to comment, as did Microsoft, IBM and Oracle.

• An in-house store inventory system, replacing a Microsoft tool.
• An in-house maintenance-management tool, replacing an IBM product. According to the presentation, AI-assisted coding was key to building it.
• Its own point-of-sale system to replace Oracle Simphony, a project that, according to unnamed sources, has been underway for several years.
• Some of the software could roll out by the end of 2027, depending on testing, and the company is reviewing “every contract and service.”

Why are so many companies rethinking SaaS right now?

For more than a decade, buying software as a service was the prudent choice: someone else built it, hosted it and kept it up to date. Three things changed that math. The first is price: Vertice, which manages software contracts, measured in January 2025 that SaaS prices had risen 11.4% in a year, against 2.7% average inflation across the G7, and in June 2026 it recorded a record 16.4%.

The second is waste. Zylo’s 2026 index, covering more than 40 million licenses, found that 36% of them go underused and that median SaaS spend has reached $9,455 per employee per year. The third, and newest, is that building is no longer the slow option: McKinsey reported in August 2026 that 32% of organizations surveyed decided against buying at least one software product or feature because they could build it with AI coding tools.

Markets have noticed. Between January 12 and February 23, 2026, the S&P software and services index fell 25% on fears that AI agents will replace entire applications. Microsoft’s own CEO, Satya Nadella, said in late 2024 that in the agent era the logic of business applications will move into the AI tier.

What did Ford, Walgreens, Fast Company, Santander, Claro Video and Profuturo teach me?

I have worked on technology projects for companies in very different industries: automotive, pharmacy retail, media, banking, entertainment and retirement funds. What follows is not inside information about any client; it is their public history. And that history keeps repeating itself: none of these companies has stopped swinging between building and buying.

• Ford: moved its SYNC platform from Microsoft to BlackBerry QNX in 2014, chose Google Cloud and Android in 2021, created Latitude AI in 2023 to develop driver assistance in-house, and in 2025 cancelled its own FNV4 electrical architecture over cost and delays. CEO Jim Farley named the root problem: when a supplier owns a module’s software, changing it requires the supplier’s permission. In 2022 Ford opened its Global Technology and Business Center in Naucalpan, near Mexico City.
• Walgreens: in 2019 it signed a seven-year partnership with Microsoft to move most of its infrastructure to Azure, and in 2020 it handed its IT operations to TCS under a ten-year contract. A clear bet on contracting technology capacity rather than building it.
• Fast Company: in 2003 its site ran on a homegrown content management system; in 2008 it moved to Drupal, an open-source platform, and years later it had to rebuild on a new architecture. Adopting a platform does not remove the duty to keep it current.
• Santander: it built Gravity in-house, the platform that moved its core banking to the cloud, and in November 2025 Santander Mexico finished migrating its entire technology infrastructure to it. The clearest example of owning what is critical and contracting what is generic: the software is theirs; the cloud it runs on is not.
• Claro Video: América Móvil bought DLA in 2011 to own its video platform and launched Claro Video in 2012; for delivery it contracts specialized pieces, such as Broadpeak’s delivery network. A hybrid model.
• Profuturo: in Mexico’s retirement fund system, part of the digital channel is shared infrastructure set by the regulator, such as AforeMóvil since 2017, and Profuturo still runs its own self-service app. What is regulated gets shared; the customer experience gets built.

Why does technological stability never seem to arrive?

After years on these projects I settled on an image that helps me explain it: companies are chasing a kind of singularity, a point of balance where their technology finally works the way it should. But that point behaves like an asymptote. The curve gets closer and closer to the line of stability and never crosses it.

There are two reasons. The first is that we don’t own the technology: when the roadmap, the pricing and the code belong to a vendor, every improvement depends on their priorities, not ours. The second is that almost no company had the resources or the in-house capacity to build and sustain its own software; building was a multi-year project with a large team and high risk, as the architecture Ford had to cancel shows.

That is changing. AI does not hand us stability, but it does close the gap: it lets us build better, with more testing and more review, and it makes your own software perfectible, because every improvement is in your hands rather than in a vendor’s request queue.

What does AI change in the decision to build, and what doesn’t it change?

It changes speed and team size. In GitHub’s 2022 controlled study, developers using Copilot finished the same task 55.8% faster, and in February 2026 Retool found that 35% of 817 teams had already replaced the functionality of at least one SaaS tool with their own, and that 78% plan to build more in 2026. What gets replaced first are internal pieces, not the core of the business, as the chart at the end of this section shows.

But it is worth separating what AI changes from what it doesn’t:

• Software still needs maintaining. Bloomberg says so in the Starbucks story itself: in-house software can cost less up front and more later, in maintenance and staff.
• AI still makes mistakes. A controlled study by METR in July 2025 found experienced developers were 19% slower with early-2025 AI tools. Method makes the difference: automated tests, human code review and small deployments.
• Building without the capacity to run it is still expensive. In May 2026 Starbucks retired an AI inventory-counting system and its stores went back to counting by hand. Building is only half the job.
• What does change is who can do it: a mid-sized company can now own its platform without standing up a hundred-engineer department.

Which industries will build their own technology first?

The ones paying for software that touches how they compete, and the ones paying for many licenses to use a few features. These are the industries where we see the most opportunity, with examples of the platform that usually makes sense to build:

• Retail and consumer goods: inventory, point of sale, loyalty and replenishment, as in the Starbucks case.
• Restaurants and coffee chains: ordering, kitchen, equipment maintenance and per-location dashboards.
• Banking, fintech and insurance: origination, collections, customer portals and product cores, like Santander’s Gravity.
• Retirement funds and wealth management: self-service, simulators and commission calculation.
• Healthcare and pharmacies: scheduling, prescriptions, records and regulated inventory control.
• Automotive and manufacturing: maintenance, quality, traceability and product software.
• Logistics and distribution: routing, warehouses, proof of delivery and customer tracking.
• Media, publishing and entertainment: content management, video, subscriptions and first-party advertising.
• Education: learning platforms, student information systems and family communication.
• Real estate and construction: property management, collections and project progress. In July 2026 a property manager was reported to have replaced Salesforce, Entrata and Yardi with AI-assisted in-house development.
• Travel and hospitality: direct bookings, guest experience and housekeeping operations.
• Professional services: law, accounting and consulting firms with their own document workflows, CRM and billing.

When should you not build your own software?

Building is not an ideology. Even the CEO of Klarna, the company most often cited for dropping Salesforce, clarified in 2025 that he doubts every company should do the same. Our rule is simple: own what sets you apart and buy what is the same for everyone.

Build when the software touches your competitive edge, when you pay for a hundred features and use five, when integrating the SaaS with your systems costs more than the license, or when your data is strategic and you want it under your control. Keep paying for the subscription when:

• The tool is generic and mature: email, office suites, video calls, payroll or tax accounting.
• The vendor’s value is the regulatory compliance it maintains for you, such as e-invoicing.
• You use almost every feature you pay for and the per-seat price stays reasonable.
• There is no one, inside or outside, who will run and improve what gets built.

What should we expect over the next few years?

Coexistence, not wholesale replacement. Gartner estimated that worldwide SaaS spending would approach $300 billion in 2025, so subscriptions are here to stay. What is growing faster is in-house software: Grand View Research expects the custom software development market to grow from $45.4 billion in 2025 to $178.9 billion in 2033, an 18.8% compound annual growth rate.

Mexico is moving the same way. According to IDC, Mexican companies will spend $776 million on AI software and services in 2026, up 52% from the year before. The question for the coming years will not be whether to use AI, but whether you use it to subscribe to more tools or to build your own.

How does Editando Ideas help a company reduce its dependence on SaaS?

Since 2014 we have designed, built and maintained software for companies in Mexico and the United States. Firms like ours will be a key piece of this transition: we bring the in-house capacity most companies lack, without them having to build an engineering department from scratch. This is how we run a project like this:

• Dependency inventory: which subscriptions you pay for, how much of each you use, what data they hold and what would happen if they raised prices or disappeared tomorrow.
• Decision map: what to keep, what to replace and what to build, prioritized by impact on your operation and by risk.
• A first in-house module with a fixed scope: usually an internal tool or a workflow, built with AI and with human review of every change.
• Migration without blackouts: the new system runs alongside the old one until it proves itself, and data is migrated with verification.
• Operation and continuous improvement: support, security, updates and new features, so what you own does not turn into technical debt.
• Real ownership: the code lives in a repository your company controls and the work belongs to you under the contract. If one day you decide to run it with your own team, we hand it over documented.

Sources

The figures in this post come from these publications; each chart links to its own source below it.

• Bloomberg News, “Starbucks Taps AI to Reduce Reliance on Microsoft, IBM Software,” July 9, 2026, as published by Fortune and, in Spanish, by Bloomberg Línea.
• Vertice, SaaS Inflation Index (January 2025) and SaaS Inflation Rate (2026).
• Zylo, 2026 SaaS Management Index (January 2026).
• McKinsey, The state of AI in 2026 (August 2026).
• Retool, 2026 Build vs Buy Report (February 2026).
• GitHub, controlled study on Copilot (2022), and METR, controlled study on AI and experienced developers (July 2025).
• Gartner, worldwide public cloud spending forecast (November 2024); Grand View Research, Custom Software Development Market (August 2026); IDC, via Mexico Business News (August 2026).
• Public history of Ford, Walgreens, Fast Company, Santander, Claro Video and Profuturo: company press releases, SEC filings and trade press.

Related services

• Technology strategy
• Technical audit
• Web platforms
• Talent
• AI, app, platform or automation?
• When to move off spreadsheets
• Monolith or microservices