EnterpriseAI Archives - Sazko Solutions

The Price of Intelligence Keeps Falling. Here’s What to Do With That

Mahesh Kumar — Thu, 06 Aug 2026 09:35:00 +0000

In August 2026, one major provider cut the price of a capable model by 80% in a single move, and another shipped a new fast model at half the previous generation’s cost. Token prices have been falling steeply for over a year, and August made it obvious this is a trend, not a promotion. The question worth asking is what your organisation should actually do differently because of it.

Falling cost changes what’s worth automating

Workflows that didn’t pencil out at last year’s prices — high-volume classification, first-draft generation across thousands of documents, always-on monitoring with an AI layer — start to make sense. It’s worth revisiting the automation ideas you shelved as “too expensive to run at scale” twelve months ago.

It also removes an excuse for sloppy architecture

When calls were expensive, there was pressure to be careful about them. Cheap calls make it tempting to spray model requests everywhere without evaluation, monitoring, or a clear reason. Cheaper inputs, more of them, still adds up — and unmonitored AI in production is a risk regardless of the bill.

Where the freed-up budget should go

Evaluation harnesses for the workflows you’re expanding.
Observability, so you know what your AI systems are actually doing.
The human review capacity that volume increases will demand.

Summary: Cheaper AI is a real opportunity, but the saving is only worth having if it funds the discipline that keeps expanded AI use trustworthy.

#AI #CostOptimization #EnterpriseAI #TechStrategy #SazkoSolutions

Published by Sazko Solutions – Driving Innovation in AI Economics and Delivery

The post The Price of Intelligence Keeps Falling. Here’s What to Do With That appeared first on Sazko Solutions.

Own the Model, Own the Trust: What Mayo Clinic’s AI Deal Says About Regulated Industries

Mahesh Kumar — Thu, 04 Jun 2026 09:40:00 +0000

In June 2026, Microsoft deployed a customised version of a frontier model inside Mayo Clinic’s own network, with the clinical model owned entirely by Mayo Clinic to protect patient records and preserve trust. The arrangement is a template a lot of regulated organisations are going to want, and it’s worth understanding why.

The trade being made

The default way to use a frontier model is to send data to the provider’s infrastructure. For a hospital, a bank or a government agency, that’s often a non-starter — the data can’t leave, full stop. This kind of arrangement inverts it: the model comes to the data, runs inside the organisation’s boundary, and the organisation owns the deployed artefact.

What it costs

This isn’t free. Running a frontier-class model in your own environment means real infrastructure, real MLOps capability, and a slower update cadence than a hosted API. You trade convenience and always-latest capability for control and data isolation.

When it’s worth it

The data genuinely cannot leave your environment for legal or contractual reasons.
The workload is core enough to justify the operational investment.
You have, or can build, the team to run it.

For everything else, a hosted model with strong contractual data terms is usually the better trade.

Summary: “The model runs where our data lives, and we own it” is becoming a standard requirement in regulated sectors. It’s achievable, but it’s an infrastructure commitment, not a procurement checkbox.

#AI #RegulatedIndustries #DataPrivacy #EnterpriseAI #SazkoSolutions

Published by Sazko Solutions – Driving Innovation in AI for Regulated Industries

The post Own the Model, Own the Trust: What Mayo Clinic’s AI Deal Says About Regulated Industries appeared first on Sazko Solutions.

Agents That Learn Between Sessions: The Move to Long-Running AI Workflows

Mahesh Kumar — Thu, 21 May 2026 11:15:00 +0000

In May 2026, Anthropic described a technique it called “dreaming” — letting an autonomous agent review its own past behaviour between sessions, spot patterns, and adjust how it works next time. It’s part of a broader industry push toward agents that handle long-running workflows in areas like coding, finance and legal work, rather than one-shot tasks. That shift changes what you have to build around an agent.

One-shot vs. long-running

A one-shot agent takes a request and returns a result; if it’s wrong, you try again. A long-running agent holds a goal across hours or days, accumulates state, and takes many actions before anyone checks the outcome. The failure modes are different: drift, compounding small errors, and actions taken on stale assumptions.

What “learning between sessions” needs from you

A durable record of what the agent did and why, that both the agent and a human can review.
Checkpoints where a person confirms direction before the agent commits to the next phase.
Clear boundaries on what the agent can change without approval, especially as it gets more autonomous.

The trust curve

Agents that improve over time are more useful and harder to reason about — the thing making decisions this week isn’t quite the thing you evaluated last month. Treat capability growth as a reason to strengthen oversight, not relax it.

Summary: Long-running, self-improving agents are a real step forward in usefulness. The operational discipline they require is a step up too, and it doesn’t come in the box.

#AI #AgenticAI #Automation #EnterpriseAI #SazkoSolutions

Published by Sazko Solutions – Driving Innovation in Agentic AI and Automation

The post Agents That Learn Between Sessions: The Move to Long-Running AI Workflows appeared first on Sazko Solutions.

When Everyone Gets a Copilot: What University-Wide AI Rollouts Teach the Rest of Us

Mahesh Kumar — Tue, 27 Jan 2026 11:00:00 +0000

In January 2026 the University of Manchester became the first university in the world to give every student and staff member access to Microsoft 365 Copilot, paired with training on responsible use. Whatever you think of the specifics, it’s a useful case study in something most organisations are about to face: what happens when an AI assistant goes from a pilot with twenty people to everyone, all at once.

The rollout is the easy part

Flipping the licence switch takes a day. The hard parts show up afterwards: people using the tool for things it’s bad at, inconsistent output quality feeding into real decisions, and a support burden nobody scoped. Pairing access with mandatory responsible-use training is the detail worth copying — the training isn’t a compliance checkbox, it’s the thing that determines whether the rollout creates value or noise.

What “responsible use” actually needs to cover

Where the tool’s output can and can’t be trusted without verification.
What data must never be pasted into it.
How to attribute and disclose AI assistance in work that others rely on.
Who to ask when the tool is confidently wrong.

The measurement problem

Most broad AI rollouts can’t answer “did this help?” six months later because nobody defined what help would look like. Before a wide deployment, pick two or three concrete workflows you expect to improve and measure those specifically. Vague productivity surveys won’t tell you anything actionable.

Summary: Universal access to an AI assistant is a change-management project wearing a software licence. The organisations that treat it that way get more out of it than the ones that just buy seats.

#AI #EnterpriseAI #ChangeManagement #TechStrategy #SazkoSolutions

Published by Sazko Solutions – Driving Innovation in AI Adoption and Enablement

The post When Everyone Gets a Copilot: What University-Wide AI Rollouts Teach the Rest of Us appeared first on Sazko Solutions.

Cheaper Reasoning, Not Bigger Models: What the January AI Shift Means for Buyers

Mahesh Kumar — Wed, 14 Jan 2026 09:30:00 +0000

The AI headlines in January 2026 marked a quiet but real change in direction. Instead of another round of “our model has more parameters,” the major labs and cloud providers spent the month talking about routing, tool use, and structured reasoning — getting more value out of each token rather than simply making models larger. For anyone budgeting for AI this year, that shift matters more than any single model release.

What actually changed

For two years the implicit promise was that capability scaled with size, and cost was the price of admission. January’s announcements leaned the other way: smarter orchestration, cheaper reasoning paths, and models that decide how much effort a question deserves. The practical result is that the same task can now cost a fraction of what it did a year ago — if the system around the model is built to take advantage of it.

Why this favours disciplined adopters

A price drop only helps you if your architecture can absorb it. Teams that hard-wired a single expensive model into every call don’t automatically benefit; they have to re-engineer. Teams that abstracted model calls behind their own interface can route the 80% of requests that don’t need a frontier model to a cheaper reasoning tier, and reserve the expensive path for the ones that do.

What to do with the savings

Re-run your cost model. Assumptions from mid-2025 are probably wrong by a wide margin.
Tier your workloads — not every request needs the same model.
Reinvest the headroom in evaluation and monitoring, which is where most AI projects are actually under-resourced.

Summary: The lesson from January isn’t “AI got cheaper” — it’s that the teams positioned to benefit are the ones who treated model choice as a variable, not a constant.

#AI #TechStrategy #EnterpriseAI #CostOptimization #SazkoSolutions

Published by Sazko Solutions – Driving Innovation in AI Strategy and Delivery

The post Cheaper Reasoning, Not Bigger Models: What the January AI Shift Means for Buyers appeared first on Sazko Solutions.