Thought leadershipQuarterly briefingQ2 2026

Stop treating AI like a smart intern.

It is no longer a sidecar for note-taking and novelty. It is becoming an execution layer — and the organizations that pull ahead over the next twelve months will use it to compress cycle time, raise execution quality and shrink the distance between a decision and the work that follows it.

9 min read

The quarter in one figure

  1. 60–70%

    of the time spent in knowledge-based roles sits in activities generative AI can automate. Not roles eliminated — time reallocated.

    McKinsey, as cited in the Q2 2026 AWALI briefing

Pace is a business problem, not a technology one.

The pace of capability change now outstrips the quarterly planning cycle. A roadmap written in January is being executed against a market that repriced twice before it shipped.

The fix is not a better roadmap. It is a shorter loop: kill the six-month pilot cycle and replace it with a monthly operating review that asks the same three questions every time.

Every thirty days · owned by operations

  1. What just became viable?

    Something that did not work ninety days ago now does. Name it, or you will find out from a competitor.

  2. What got cheaper?

    Price moves are the fastest-moving variable in this market and the easiest one to bank.

  3. What should be redesigned now?

    Given the first two answers, which workflow is now worth rebuilding rather than tuning?

Then ask them again in thirty days. A company without this rhythm is not learning slower than its competitors by a little.

The cadence matters more than the answers. A company running this monthly absorbs four times as many capability changes per year as one running it quarterly, from exactly the same market.

So whatIf you do not have this rhythm, your company is not learning too slowly relative to some ideal. It is learning too slowly relative to whoever does have it.

Cost is collapsing. Re-optimize what you already built.

Any AI workflow built six to twelve months ago is almost certainly overspending, often by a multiple rather than a margin. The prices underneath it have moved and the workflow has not.

This is the least glamorous item on the list and the fastest to bank: an audit of what you already run, priced against what the same work costs today.

The audit, in four moves

  1. Audit what you already run

    Document processing, voice, research, analytics. Anything built against a price list that is now out of date.

  2. Swap in better-priced models

    The cheapest model that clears your quality bar is the correct model, and it is rarely the one you started on.

  3. Consider local models

    For confidential or high-volume workloads, weights you hold change both the economics and the exposure.

  4. Reallocate, do not just save

    The point of the saving is to fund the next redesign, not to show up once in an operating review.

So whatEvery dollar recovered here drops straight to the bottom line and needs no new capability, no new vendor and no change management. It is the closest thing to free money in this brief.

Agentic AI is execution, not answers.

The shift is from "AI gives me an answer" to "AI gets the work done". That is a different procurement question, a different risk profile and a different definition of success.

Start where the work is structured, repetitive and multi-step, and where the cost of a wrong step is recoverable — research, briefing preparation, browser-based data gathering, internal QA. Bounded automation only, with lanes drawn before anything runs.

So whatThe constraint on agentic value right now is not model capability. It is how clearly you can describe the boundaries of a job — and most organizations have never had to write that down.

Memory and context are the strategic asset.

The base model is a commodity. Your business context is not, and it is the only input to the system that your competitor cannot also buy.

Value compounds through three things specifically: clean knowledge sources, memory that persists across sessions rather than restarting every morning, and context shaped to the role of the person asking.

So whatGeneric AI is available to everyone, including the competitor you worry about. AI that knows your customers, your workflows, your exceptions and your past decisions is the part that separates you — and it is built, not bought.

Stop standardizing on one model.

Two things arrived at once this quarter. Image, video and voice are production-ready rather than demo-ware — training content, sales visuals and onboarding assets now take hours instead of weeks. And the argument about which foundation model to standardize on stopped being worth having.

There is no best model. There is a best model for this workflow, at an acceptable risk, at the right cost — and the answer differs three times inside the same company.

If you are evaluating more than three tools at once, you have already lost focus. The goal is not to adopt everything. It is to build a disciplined cadence for absorbing useful capability into the business.

So whatEvaluate workflow by workflow rather than issuing a standard. And where no available tool fits the workflow properly, building the tool is now a reasonable answer rather than an ambitious one.

The five-point plan

Five things to put in motion this quarter.

Each one names the executive who owns it, how long it should take to stand up, and the target to hold it to. A step with no owner and no finish line is a wish.

  1. Install a 30-day AI operating review

    Ask monthly what became newly viable, what got cheaper, and which workflow should be redesigned now. Without this rhythm the company is learning too slowly.

    Owner
    COO or Head of Operations
    Time
    2 weeks to stand up
    Target
    Two to three workflows improved per month on cost, time or quality

  2. Audit existing AI workflows for cost

    If it was built more than a few months ago it can probably run materially cheaper today. Audit it, swap the model, reallocate the saving.

    Owner
    CFO with CTO
    Time
    30 days to a full audit and reallocation plan
    Target
    A 20–40% reduction in AI and tooling spend with no loss of performance

  3. Build internal tools where speed creates advantage

    Custom tools are now faster to build than most SaaS is to buy and configure. Evaluate on return, and give each one an internal champion.

    Owner
    CTO or Head of Product
    Time
    2–4 weeks per tool for initial builds
    Target
    A two- to fivefold reduction in cycle time on the targeted workflow

  4. Pilot one bounded agentic workflow

    Start where execution is repetitive and risk is recoverable — research, briefing prep, browser-based data gathering, internal QA. Clear bounds, one workflow.

    Owner
    CTO, Head of Automation or Ops Lead
    Time
    2–3 weeks to first deployment
    Target
    Half the manual effort removed from the selected workflow, with no uncaught errors

  5. Treat memory and context as a strategic asset

    Generic AI is available to everyone. AI that knows your customers, workflows, exceptions and decisions is what starts to set you apart.

    Owner
    CTO or Data Lead
    Time
    4–6 weeks for initial ingestion, structuring and deployment
    Target
    Above 90% accuracy on internal queries, and a measurable fall in time-to-answer

Targets to set, not results observed. These are the commitments a leadership team makes when it starts; none of them are AWALI measurements or client outcomes.

Where the new capability is worth a bet.

Sorted by how ready each one is rather than how interesting it is. The goal is not to adopt everything — it is to have a disciplined cadence for absorbing the useful parts.

Capability bets, by readiness

Tier 1 · immediate return

Available now, obvious payback, low change cost.

  • Presentation and document generationProduction-quality decks, white papers and web output from structured input.
  • Agentic researchCited research and browser-based execution rather than a single-shot answer.
  • AI-assisted engineeringCompressing internal build and automation cycles for teams that already ship.
  • An internal knowledge layerSearch, retrieval and context across company data — the foundation the rest sits on.

Tier 2 · high potential

Real but early. Pilot rather than roll out.

  • Agentic workflow orchestrationCoordinating multi-step work across tools and systems.
  • Scheduled and background executionWork that runs on a clock or a trigger rather than on a prompt.
  • AI-native app buildingFull-stack generation of internal applications by non-engineering teams.
  • AI inside existing productivity toolsCapability that arrives where people already work rather than in a new tab.

Tier 3 · specific use cases

Worth knowing about. Not worth a program yet.

  • Ultra-fast inference hardwareWhere latency, not capability, is the binding constraint.
  • Local and open-weight media modelsFor confidential workflows that cannot leave the building.
  • Next-generation accelerator architecturesChanges the economics again, on a longer clock.
  • Deep multimodal reasoningFor analysis that spans documents, images and structured data at once.

Illustrative, as of Q2 2026. Named products are examples of a category, not endorsements, partnerships or performance claims — and a list like this dates quickly.

Media — fast-moving areas to experiment in

Image and scene

  • Context-aware generationHigh-fidelity images and scenes from minimal input.
  • Product and lifestyle imageryRemoves the photoshoot from a large share of catalog work.

Video

  • Image-to-video and text-to-videoRapid generation for short-form and internal communication.
  • Persistent scenes and continuityThe emerging capability that makes narrative video usable.
  • Avatar presentersSales, training and internal messaging without a studio booking.

Audio

  • Voice and narrationCloned or synthetic voice at production quality.
  • Music and soundtrackCustom scoring from a prompt, licensed cleanly.

Illustrative, as of Q2 2026. Named products are examples of a category, not endorsements, partnerships or performance claims — and a list like this dates quickly.

AI is getting faster, cheaper, more local, more agentic and more operationally useful — all at once. The organizations that win will not have the best prompts. They will have redesigned work faster than their competitors did.