Signal #94: Cognizant Bought an AI-Native Dev Shop

You've spent years close to a broken industry or process. You know exactly what would fix it. But you’ve never had the time or resources to make it happen.

With Lovable, your expertise is all you need. Describe the solution to your problem — a website, a dashboard, an internal tool — and get professional-grade software back. You make every call on how it works, because you're the one who knows what it needs to do.

Start today and take $5 off your first month.

Build the fix, $5 off

_{^{Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.}}

Cognizant bought a nearshore shop that sells itself as AI-native while EPAM forecast a flat quarter. Frontier model prices fell again, Croatia put €932M behind AI, and Brussels started asking where your engineers sit. Buyers are paying for teams that deliver with AI, not for headcount.

Hello, friends.

Two weeks ago I put Zani on the App Store: an iPhone app that turns Albanian meetings and long WhatsApp voice notes, Gheg included, into notes with decisions and owners. Claude Code wrote most of it from a spec, screen designs and a page of rules I gave it. The code was the easy part. Gheg was hard: one speech model turned "U prish 2 herë" (it broke twice) into "both of them broke", which reads fine and means something else.

Give it a test drive and let me know what you think.

On 6 November Cognizant closed a deal for Wizeline, a nearshore engineering firm that describes itself as AI-native. A day earlier EPAM, the biggest IT services company to come out of Eastern Europe, told investors to expect roughly flat revenue next quarter, with about 63,650 people on the payroll.

I think the big buyers have stopped paying a premium for headcount. They pay for small teams that already deliver with AI and can prove it. Zani taught me the other half: the agent writes the code, but knowing the language and the user is the part nobody can copy. That's good news for a 40-person shop in Prishtina or Tirana that has the proof. It's bad news for anyone still selling people by the month.

The model prices under your fixed quotes moved again this week, and whoever re-prices first keeps the margin. Croatia put a €932 million AI plan on the table with projects a regional team could bid for, and the comment window shuts on 15 November. And Brussels is drafting cloud rules that ask where your engineers sit, which puts Kosovo and Albania in the same box as London.

My worry is the label. "AI-native" is turning into what "agile" was ten years ago, a word every agency puts on its homepage. Buyers will start asking for numbers. I'd rather we had them ready.

Dev-shop founders, read the Cognizant story and the pricing story together. Forward the Croatia piece to whoever handles your public-sector bids.

On 6 November Cognizant closed its purchase of Wizeline, a nearshore engineering firm founded in 2013 that builds digital products for US companies in media, finance and healthcare. Wizeline calls itself an "AI-native" product engineering company. The price wasn't disclosed. A day earlier EPAM reported revenue up 2.3% for the quarter and guided the next one to between 0.2% down and 0.9% up, on about 63,650 staff.

The big integrators aren't buying more engineers. They're buying teams that already ship with AI, because growing by headcount has stopped working for them. EPAM launched a "Frontier AI Services" line in the same quarter, which tells you where it expects growth to come from. The fair objection is that one undisclosed deal and one flat quarter don't make a trend, and Cognizant may just have wanted Wizeline's media clients. Maybe. But when a buyer pays for a firm's AI story, it's telling every other seller what it values.

Write one page that answers the question an acquirer or a new client will ask: how much of your delivery runs through AI, and what did that change in speed or price? Back it with two client examples. Have it ready before your next sales trip to Munich or London.

Croatia's draft national AI plan sets aside €932 million to 2032, about €382 million of it in the first action plan to 2028. Around €115 million goes to bringing Croatian companies into strategic AI projects, €36 million of that for startups and scaleups. Some of it is work you could bid for: €6 million for an AI assistant on the e-Građani citizen portal, €2.6 million for a Croatian-language model and €39 million for a national AI factory, plus €800,000 in AI vouchers for small firms.

A state AI budget is the nearest paying client most Balkan dev shops will get this decade, and the rules for spending this one are being written now. Netokracija notes that the 133-member working group leaned on ministries, agencies and universities, and its host argued that companies building AI every day deserved a bigger say. That's how tenders end up shaped around the institutions that drafted them. Plans like this often slip, and €800,000 in vouchers across a whole country is small. So read it for the projects, not the vouchers.

The plan is open for comments on e-Savjetovanja until 15 November. Selling into Croatia, or partnered with a Croatian firm? File one comment on how the e-Građani agent and the language model should be tendered. When Kosovo or Albania write their own plan, this is the one to compare it against.

On 9 November Anthropic released Claude Fable 5.2 at $8 per million input tokens and $40 per million output, down from $10 and $50 for Fable 5.1, with batch jobs at half that. It scores 69.9% on the FrontierSWE v2 coding test, up from 56.3% for the previous version. The same day Meta released Muse Ember, a small model for the steps an agent repeats all day, at $0.10 per million input tokens. Meta's own summary: "Muse Wonder plans, Muse Ember handles the volume."

Every fixed-price quote you wrote this year assumed a model cost that is now out of date, and the margin goes to whoever re-prices first. Anthropic's own advice is to start most work on Opus 5.5, move up to Fable for the hardest jobs and down to smaller models when the task allows. That's routing, and it's where a 30-person shop can beat a 3,000-person one, because a small team can change its stack in a week. The weak spot: Meta hasn't published Ember's output price or its scores, so treat the ten cents as a claim until you've tested it.

Take one live client project this week and price its AI bill three ways: everything on one frontier model, a planner plus a cheap model doing the volume, and batch for anything that isn't urgent. Then decide whether the saving goes into your margin or into a lower bid.

The EU's proposed Cloud and AI Development Act would sort cloud and AI providers bidding for public contracts into four "assurance levels". At the top two, ITPro reports, staff must be based in the EU, data must stay in the EU and providers under third-country control are out. UK trade body techUK puts the cost to British firms at £19.6 billion to £22.3 billion over five years and wants changes before the UK-EU summit on 20 November.

For a dev shop in Prishtina or Tirana, "third country" is the same label as for London, and your clients' public bids will start asking where your people are. The Commission says only about 10% of public procurement would fall under the top two levels, so most work stays open. But techUK's worse case assumes private buyers copy the rules. That's how checklists spread: a bank or insurer adopts a public standard because it's easier than writing its own.

Ask your three biggest EU clients whether they bid for public cloud or AI work, and at which level. Where the answer is yes, work out now which projects could run with EU-based staff or an EU partner, before it turns up as a clause in a renewal.

Sales: Otter is adding 27 languages to its meeting notes. Otter added 13 languages that work today, including German, Dutch, Polish and Turkish. Romanian, Czech, Hungarian, Greek, Macedonian and the Nordic languages are listed as coming soon. Albanian and Serbian aren't on the list. Run it on your next German or Dutch sales call and compare the notes with your own.

Marketing: Medium's TK is a free AI editor. TK puts an assistant inside the editor: type "@TK" for headline feedback, a style-guide check or a quick fact. It can publish a note on how AI was used, and it connects to Claude through an MCP server or a skill. It's web only for now. Draft your next case study there and keep the AI note.

Dev: Automattic's Spacefast gives vibe-coded apps a backend. The WordPress.com company launched Spacefast with databases, login, file uploads, email and analytics. It connects to Claude, ChatGPT, Cursor and other agents, or to a Git repo. Pricing isn't public yet. Try it on the next client prototype that needs a login before it needs a real backend.

Compliance: the AI Act's content-marking deadline is 2 December. The digital omnibus delayed the high-risk rules but left this one standing: generative systems on the EU market before 2 August 2026 need machine-readable marks on what they generate by 2 December. List every client product with a text, image or audio generator, and check what it embeds today.

Running a dev team that sells abroad? Hit reply and tell me the first question clients now ask about how you use AI.

Next edition soon,
Çelik