If 2023–2024 were the “try AI everywhere” years, 2025 is the year the messy experiments start turning into repeatable playbooks. Boards want outcomes, buyers want proof, and teams want the fastest path from “we could” to “we did.” You’ll see the stack get simpler up top (passkeys, agent bundles, ready-made data patterns) and more interesting underneath (vector-native storage, FinOps for AI, new pricing knobs). Below are the moves we’re watching—and the one-sentence nudge on what to do next for each.
1) From copilots to agents (with human guardrails)
The assistant that lived in a sidebar is growing up into task-orchestrating agents—small pieces of software that can follow multi-step workflows, use company data, and ask for approval before doing anything risky. Microsoft is shipping an Agent Store inside M365 Copilot and adding multi-agent orchestration so these helpers can work together across apps like Teams, Word, and Excel.
Meanwhile, OpenAI leadership is openly talking about “millions of cloud-run agents” working under human supervision in the near future.
What to do now: pick one supervised workflow (support triage, finance ops, growth ops), log every action, and require human approval for anything financial, legal, or privacy-sensitive.
Sources: Microsoft Agent Store, Build 2025 multi-agent orchestration; OpenAI’s agent outlook. (Microsoft for Developers)
2) AI governance stops being optional (EU AI Act reality)
Buyers will increasingly ask about model transparency, data provenance, and risk class. That’s not just diligence—it’s the EU AI Act timeline playing out: prohibitions and AI literacy duties kicked in Feb 2, 2025, GPAI obligations began Aug 2, 2025, and most other obligations ramp by Aug 2026.
What to do now: publish a lightweight model & data overview, keep auditable evals, and document retention/opt-outs.
Sources: EU AI Act timeline (official pages + Parliament brief). (Digital Strategy)
3) FinOps for AI gets teeth
Cloud costs are rising—and AI workloads amplify that. Flexera’s 2025 report pegs managing cloud spend as the top challenge. The FinOps Foundation is now formalizing FinOps for AI (unit economics for tokens/inference, show-back, carbon tracking).
What to do now: meter cost per 1K tokens / per query / per workflow, set spend SLOs, and separate Cloud vs AI scopes in dashboards.
Sources: Flexera 2025; FinOps for AI overview. (Flexera)
4) Passkeys become table stakes
Password UX finally improves at scale. Real-world rollouts show passkeys deliver faster sign-ins and fewer errors; Google benchmarked passkeys as faster and less error-prone than passwords, and FIDO’s 2025 initiatives point to broad, real deployments. Microsoft and others are pushing passwordless defaults across consumer and enterprise surfaces.
What to do now: offer passkeys alongside SSO and make passwordless the preferred path; measure completion and recovery rates.
Sources: Google Security Blog (passkeys), FIDO Alliance 2025 update; Microsoft security update. (Google Online Security Blog)
5) The vector-native data layer goes mainstream
As semantic search, recommendations, and RAG become standard features, the data tier is shifting: clouds and databases keep adding vector search natively (e.g., Google Cloud expanding vector capabilities across Bigtable, Cloud SQL, Firestore, Spanner), while editorial coverage tracks vectors’ rise from niche to default.
What to do now: start with a curated corpus and retrieval evals before scaling; treat embeddings like sensitive data (region, retention, encryption).
Sources: InfoWorld on vectors & Google Cloud updates; WSJ backgrounder on vector DBs. (InfoWorld)
6) Pricing evolves: seat + usage (and AI meters)
AI moves value from “who can log in” to “how much gets done.” Expect hybrid models—base seats for collaboration plus metered usage (tokens, tasks, automation runs). OpenView’s research remains the canonical primer, and 2025 surveys (Metronome) show continued momentum toward usage-based and hybrid pricing.
What to do now: meter the one or two activities most correlated to value, publish guardrails (quotas, caps), and ship an admin usage dashboard.
Sources: OpenView on UBP; Metronome’s 2025 report. (OpenView)
7) Suite-native AI shortens time-to-value
Big platforms are bundling ready-to-run agents for sales, service, marketing experiments, etc., so teams click into outcomes without wiring every integration themselves. Microsoft’s Copilot updates are a clear signal: build, publish, and discover agents in-suite.
What to do now: align to major suites’ event schemas and ship prebuilt playbooks that work out-of-the-box.
Sources: Microsoft Agent Store (developer blog). (Microsoft for Developers)
8) Buyers want proof, not demos
Enterprise buyers will keep asking for before/after metrics and deployment friction numbers. 2025 CIO surveys show AI is the top tech priority, but spend is tied to clear ROI and faster paths to value.
What to do now: publish reference playbooks (“save X hours,” “lift conversion Y%”), include sample data, and offer sandbox orgs for week-one pilots.
Sources: CIO.com’s State of the CIO 2025. (CIO)
9) Desktop-as-a-Service (DaaS) creeps into the mainstream
Between AI-heavy workloads, device management overhead, and security constraints, a slice of teams are moving to DaaS and thin clients. Analysts and industry write-ups point to improving total cost of ownership and a steadily growing market through 2029; vendors are jostling in the Gartner MQ.
What to do now: if your product is resource-intensive, validate on major DaaS stacks and document “known-good” profiles.
Sources: TechRadar Pro on DaaS cost shifts; Virtualization Review & AWS on 2025 MQ. (TechRadar)
10) Macro tailwind: cloud + AI budgets keep growing
Big picture: public cloud spend is forecast around $723B in 2025, and overall IT spend is projected north of $5.4T, driven by AI infrastructure and software. Translation: budgets exist—but scrutiny does too.
What to do now: tie your roadmap to concrete business outcomes and instrument usage & value so procurement can defend the spend.
Sources: CIO Dive on cloud spend; TechRadar Pro & others on total IT spend. (CIO Dive)
A tiny 2025 checklist (copy/paste into your sprint board)
- Ship one supervised agent where you already have context & permissions. (Microsoft)
- Add passkeys, make passwordless the default path. (Google Online Security Blog)
- Stand up FinOps for AI: unit costs, show-back, spend SLOs. (FinOps Foundation)
- Pick a vector store and write retrieval evals before scaling RAG. (InfoWorld)
- Revisit pricing: meter what maps to value; show live usage to admins. (Metronome)
- Publish governance basics (model & data notes, retention, provenance). (Digital Strategy)
Sources:
- Microsoft: Agent Store and Build 2025 multi-agent orchestration. (Microsoft for Developers)
- OpenAI agents outlook: Business Insider. (Business Insider)
- EU AI Act timeline: EU Commission & European Parliament brief. (Digital Strategy)
- FinOps for AI: FinOps Foundation + Flexera 2025 cloud report. (FinOps Foundation)
- Passkeys: Google Security Blog, FIDO Alliance 2025. (Google Online Security Blog)
- Vector data: InfoWorld on vectors & DB updates, WSJ explainer. (InfoWorld)
- Pricing: OpenView UBP, Metronome 2025. (OpenView)
- Macro spend: CIO Dive on $723B cloud, TechRadar Pro on $5.4T IT. (CIO Dive)
Leave a Reply