AWALI Thought leadershipQ4 2026
Sources
Skip the intro

AWALI · Q4 2026 briefing for CEOs and executive teams

  1. AI got ~12× more capable.

    Longer autonomous task horizon

    The length of software task frontier models complete at 50% success grew from about 1 hour to about 12 hours in a year.

  2. Equivalent performance got ~13× cheaper.

    Cheaper each year to reach the same AI performance

    The cost of reaching a fixed benchmark score falls about 13× a year when users switch to the cheapest capable model.

  3. Nearly 9 in 10 organizations now use AI somewhere in the business.

    Adoption is approaching ubiquity.

    Share of organizations using AI in at least one business function, not necessarily at scale.

  4. Yet only 37% report any EBIT impact.

    EBIT impact isn't.

    Slightly fewer than a year earlier, and most who report it attribute less than 5% of EBIT to AI.

The technology is no longer the constraint.The enterprise is.

Intelligence is becoming abundant. Execution is not.

Q4 2026

The enterprise becomes programmable.

One operating system for AI. Each of the eight changes reshapes one layer.

Governance surrounds it, enforced while it runs. Identity · Authority · Permissions · Observability · Reversibility

8 enterprise changes. 6 leadership decisions.

The summary takes about five minutes. Explore any change for the evidence.

Start the briefing

Conceived, researched, written and built by Scott ThomasAWALI · 65 sources · 44 developments tracked

The programmable enterprise: models, context, decisions, loops, workflows, voice, software and creative interfaces, and execution, surrounded by governance: identity, authority, permissions, observability and reversibility.
Next briefingOne email a quarter
Next briefing: sign up at the end

The executive map

8 changes. 6 decisions.

Six call for a leadership decision this quarter. Two are watch areas. Each one feeds your 90-day agenda.

  1. 01Intelligence becomes infrastructureThe best AI model depends on the task, not on the leaderboard. 11×difference in cost between the lowest and highest effort settings of a single frontier modelBenchmarkArtificial Analysis Decision 1Set up task-level model routing
  2. 02From loops to graphsPrompts improve an interaction. Loops automate a job. Graphs redesign a workflow. 19→44%accuracy after a test-and-retry step was added, with no change of modelResearchRidnik et al. · Jan 2024 Decision 2Redesign one cross-functional workflow
  3. 03Routine decisions can run on AIAutomation creates capacity. What leaders do with it creates value. $0.02to process 1,000 routine decisions with a purpose-built decision model (TypeSafe's Jev), at list priceAWALI calculation · TypeSafe, OpenAI & Anthropic list prices Decision 3Decide where AI-created capacity goes
  4. 04The interface recedesVoice is becoming a practical way to get work done. 125,000+businesses use Wispr Flow for voice dictation, including most of the Fortune 500PressPulse 2.0 · Aug 2026 Decision 4Choose one transaction to run by voice
  5. 05Custom software becomes rational againCustom software deserves a fresh test against SaaS. 32%of organizations have dropped at least one software purchase because they built the capability themselvesSurveyMcKinsey & Company · Aug 2026 Decision 5Re-evaluate one differentiated workflow for build versus buy
  6. 06Creative becomes a production systemCreative AI is becoming a production system. ~$500Mreported annual revenue run rate at Higgsfield, which licenses most of its models from othersReported, unverifiedTechTimes · Jun 2026 WatchBaseline creative testing speed
  7. 07AI is changing faster than annual plans can respond44 significant AI developments in nine months. ~4.3monthsfor the length of task AI can complete to double, based on independent measurementBenchmarkMETR · Jan 2026 WatchInstrument your AI planning cadence
  8. 08AI agents need limits built inAgent risk is about access more than intelligence. 34%of organizations apply the same security controls to AI agents as to employeesSurveyOkta · May 2026 Decision 6Set real-time controls for AI agents

01Change 01 · ModelsIntelligence becomes infrastructure

The best AI model depends on the task, not on the leaderboard.

The fact

11×

cost gap between the cheapest and most expensive settings of one frontier model

BenchmarkArtificial Analysis
Why it matters

Model strategy is becoming an economics and routing discipline.

Decision 1 of 6

Set up task-level model routing

Owner CTO

Route five real tasks to the right model · plus the evidence and how to start

Figure 01

Five tasks, four models, three providers

Choose a task to see which model it goes to and what it costs.

AWALI scenario · routing policy on Artificial Analysis dataBenchmarkArtificial Analysis
AWALI illustrative routing policy applied to the Q4 Artificial Analysis snapshot. Top option: Claude Opus 5.5 (max), $5.98 per task.
TaskComplexity / riskLatency sensitivityCost sensitivityQuality barRouted toIndexCost per taskCheaper than top option
Classify support tickets1 / 2highhigh22.5GPT-6 Luna · medium30$0.020299×
Real-time voice agent2 / 4highmed34.5MiMo-V2.6-Pro · default46$0.1346×
Review supplier contracts4 / 4lowhigh48.5GPT-6.1 Sol · high50$0.3219×
Draft a board memo4 / 5lowlow53.0Claude Opus 5.5 · high54$1.823×
Fix a production bug5 / 5lowlow60.0Claude Opus 5.5 · max (best available)58$5.98—
Artificial Analysis Intelligence Index v4.3.2 · Cost per index task (USD) · as of 2 Oct 2026.
ModelProviderOpen weightsEffortIndexCost per task
GPT-6 LunaOpenAINolow22$0.0045
GPT-6 LunaOpenAINonone18$0.010
GPT-6 LunaOpenAINomedium30$0.020
GPT-6 LunaOpenAINohigh33$0.030
GPT-6 LunaOpenAINoxhigh35$0.040
GPT-6 LunaOpenAINomax38$0.070
GPT-6.1 SolOpenAINolow42$0.13
MiMo-V2.6-ProXiaomiYesdefault46$0.13
DeepSeek V4.1 FlashDeepSeekYesnone25$0.15
GPT-6.1 SolOpenAINomedium48$0.21
GLM-5.3-FlashZ AIYesdefault42$0.25
DeepSeek V4.1 FlashDeepSeekYesmax39$0.27
GPT-6.1 SolOpenAINohigh50$0.32
GPT-6.1 SolOpenAINoxhigh51$0.39
Claude Opus 5.5AnthropicNolow42$0.55
Claude Sonnet 5.5AnthropicNomedium41$0.59
GPT-6.1 SolOpenAINomax52$0.72
GPT-6 AstraOpenAINolow46$0.82
GLM-5.3Z AIYeslow34$0.85
Gemini 3.8 FlashGoogleNomedium40$0.93
Claude Sonnet 5.5AnthropicNohigh47$1.08
Kimi K3KimiYeslow30$1.15
Gemini 3.8 FlashGoogleNohigh41$1.24
Claude Opus 5.5AnthropicNomedium51$1.34
GPT-6 AstraOpenAINomedium50$1.54
GPT-6 AstraOpenAINohigh51$1.73
Claude Opus 5.5AnthropicNohigh54$1.82
Gemini 4 ArgonGoogleNohigh53$1.99
Kimi K3KimiYesmax44$2.00
GLM-5.3Z AIYesmax45$2.01
GPT-6 AstraOpenAINoxhigh52$2.31
Claude Sonnet 5.5AnthropicNoxhigh52$2.74
GPT-6 AstraOpenAINomax53$3.26
Claude Opus 5.5AnthropicNoxhigh56$3.46
Claude Opus 5.5AnthropicNomax58$5.98
Claude Sonnet 5.5AnthropicNomax56$7.62
Gemini 3.8 FlashGoogleNolow33not published
The explanation

Sending every task to the top model can cost up to 299× more than routing each one to the cheapest model that is good enough.

What this means

Model strategy is becoming a routing and economics discipline. A single standard model tends to overpay for routine work and underperform on complex work, and at enterprise volumes that gap can be a large share of the AI budget.

Decision 1 of 6 · how to start

Set up task-level model routing

For your three highest-volume AI workflows, define the quality each one requires. Then test at least two providers, three effort settings and one open-weight model, using your own data. Review the routing every month.

Suggested owner: CTO
The research

Two trends are running at the same time. Epoch AI estimates that reaching a given level of AI performance gets about 13 times cheaper each year. A separate academic study finds that running the very newest models keeps getting more expensive. Routing is how a company captures the first trend without paying for the second.

A single model's cost can vary by up to 15 times depending on how much reasoning effort it is allowed to use. Often the cheapest option that meets the quality requirement is a frontier model at a low setting, or a smaller model at a high one.

Open-weight models now score about 12 points below the top of the benchmark at a small fraction of the price. MiMo-V2.6-Pro, for example, scores 46 at $0.13 per task, compared with 58 at $5.98 for the top setting. Because these models can run on your own infrastructure, they also address most data-residency concerns and make costs more predictable at high volume. On Vercel's gateway in June, they handled 29% of traffic for less than 4% of spending.

The data behind Figure 01

02Change 02 · Loops and workflowsFrom loops to graphs

Prompts improve an interaction. Loops automate a job. Graphs redesign a workflow.

The fact

19→44%

accuracy after adding a test-and-retry step, with the same model

ResearchRidnik et al. · Jan 2024
Why it matters

Most of the reliability gain comes from the loop. Connect loops into larger workflows only once each one works.

Decision 2 of 6

Redesign one cross-functional workflow

Owner COO

Follow one invoice from prompt to workflow · plus the evidence and how to start

Figure 02

As capability increases, so does complexity

The same disputed invoice, handled three ways. Keep scrolling to add each layer.

Vendor-reportedAnthropic · Dec 2024ResearchRidnik et al. · Jan 2024
Diagram facts for the illustrative invoice-dispute workflow.
PromptLoopGraph
What it doesImproves one interactionAutomates a jobRedesigns a workflow
Model calls11 + retries1 + retries per loop
Checks01, with a stop condition2 loops
Systems and records003
People001
Permission gates001
The explanation

Most of the reliability gain comes from one step: the system checks its own work and tries again when the check fails. In one study, that step alone raised accuracy from 19% to 44% using the same model.

What this means

Much of the reliability gain comes from the loop, not from the workflow around it. If loops are connected into a larger workflow before each one works reliably, their errors compound.

Decision 2 of 6 · how to start

Redesign one cross-functional workflow

Choose one workflow that spans at least two teams. Make sure each automated step checks its own output against a written definition of a correct result, and only then connect the steps into a single workflow.

Suggested owner: COO
The research

A prompt produces an answer. A loop checks that answer against a reliable reference, such as the purchase order, the contract or a test, and tries again until it passes or reaches a set limit. A workflow, sometimes called a graph, connects several loops with systems of record, approvals and people.

The evidence supports building in that order. In one coding study, adding a test-and-retry loop raised accuracy from 19% to 44% with the same model. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027 because of cost, unclear value or weak risk controls.

Nearly three quarters of the organizations getting the most value from AI have redesigned their workflows. Among all other organizations, about a quarter have.

The data behind Figure 02

03Change 03 · DecisionsRoutine decisions can run on AI

Automation creates capacity. What leaders do with it creates value.

The fact

$0.02

to process 1,000 routine decisions with a purpose-built decision model

AWALI calculation · TypeSafe, OpenAI & Anthropic list prices
Why it matters

Time saved is not value created. Leaders decide whether released capacity becomes savings, scale or new value.

Decision 3 of 6

Decide where AI-created capacity goes

Owner COO

See what happens to an 83‑person team · plus the evidence and how to start

Figure 03

What happens to an 83‑person invoice-review team

They check 10,000 invoices a day. Set how sure the AI must be before it decides on its own.

AWALI calculation · TypeSafe, OpenAI & Anthropic list pricesVendor-reportedTypeSafe · Sep 2026Price listOpenAIPrice listAnthropic
Per day, 10,000 decisions. All manual: 667 review hours, $40,000 labor. Illustrative: confidence profiles, 4 minutes per review, $60 per hour loaded.
DecisionThresholdAutomatedTo a personReview hoursReview laborLikely wrong
Invoice match97%7,9102,090139$8,36018
Invoice match90%8,7451,25584$5,02167
Invoice match80%9,26573549$2,941143
Ticket routing97%6,2573,743250$14,97227
Ticket routing90%7,6472,353157$9,411109
Ticket routing80%8,5841,41694$5,665245
Churn risk97%3,9156,085406$24,34129
Churn risk90%5,8484,152277$16,608146
Churn risk80%7,3682,632175$10,528368
AWALI calculation at list prices: 500 input and 20 output tokens per decision.
ModelList price per 1M tokens (in / out)Per 1,000 decisionsPer daySource
TypeSafe Jev$0.042 / $0.00$0.021$0.21TypeSafe · Sep 2026
GPT-6 Luna$0.100 / $0.50$0.060$0.60OpenAI
Claude Opus 5.5$4.000 / $20.00$2.400$24.00Anthropic

Time saved is not value created.

AI creates capacity. Value depends on what the enterprise does with it.

The technology makes the first decision. Leadership makes the next one.

Capacity released73 peopleof the 83‑person team above
  1. ReduceLower cost
  2. AbsorbHandle more volume
  3. ElevateMove people to higher-value work
  4. CreateBuild new revenue and capabilities

Company-reported. Neither company attributes these results to AI alone.

The explanation

Routine decisions now cost about two cents per thousand. The real output is the hours released, and leaders decide what they become.

What this means
  1. First-order return

    Do the same work with less.

  2. Second-order return

    Do more with the same.

  3. Strategic return

    Do something you couldn't do before.

Decision 3 of 6 · how to start

Decide where AI-created capacity goes

Choose one high-volume process. Automate the routine work. Measure the capacity released. Decide where it goes.

  • Reduce cost
  • Absorb growth
  • Elevate human work
  • Create new value
Suggested owner: COO
The research

At list prices, 1,000 short decisions (about 500 tokens of input each) cost $0.02 with Jev, $0.06 with GPT-6 Luna, $0.60 with Claude Haiku 4.5 and $2.40 with Claude Opus 5.5. These prices say nothing about accuracy, which has to be measured on your own cases.

Accuracy depends largely on how the decision is set up. In an independent test, Jev was 62.6% accurate when asked a single question and 95.0% accurate when the same decision was broken into five questions weighted using labeled data. Test on 1,000 to 2,000 past decisions and set a separate threshold for each question; where to set it is a business decision about acceptable error, to be documented and reviewed like a credit limit.

IKEA's largest retailer, Ingka, reports that Billie handled 47% of customer queries to its call centres over two years. Over the same period it trained 8,500 call-centre workers as remote interior design advisers, a channel that brought in €1.3 billion in financial year 2022 (3.3% of revenue), with a goal of 10% by 2028. Verizon gave its 28,000 service reps a Gemini-based assistant from July 2024; its consumer CEO says sales through the service team rose nearly 40% after deployment. Neither company attributes these results to AI alone.

The data behind Figure 03

04Change 04 · VoiceThe interface recedes

Voice is becoming a practical way to get work done.

The fact

~1.2s

for the leading speech models to start responding

BenchmarkArtificial Analysis
Why it matters

The value is not the speech but the systems one request can update. Voice pays off where those connections exist.

Decision 4 of 6

Choose one transaction to run by voice

Owner Chief Customer Officer

Watch one spoken request update company systems · plus the evidence and how to start

Figure 04

From a spoken request to a completed transaction

One spoken request, traced to updated company systems. A simulation; no microphone needed.

BenchmarkArtificial AnalysisPrice listOpenAI · Sep 2026
Simulated request: "Move tomorrow's client meeting to the afternoon and send them the updated prep package." Timings are representative, not measured.
StepStageStart (ms)End (ms)What happens
ListenTranscription03500Audio streams in while the person speaks.
Transcribe (streaming)Transcription2003600Speech becomes text as it arrives, so the system can start before the sentence ends.
Understand intentConversation36003780The sentence is turned into two specific actions, each with the details needed to carry it out.
Find the right recordsConversation37803990The calendar, the CRM and the document library are searched, so ‘the client’ and ‘the prep package’ are matched to real records.
Check permissionsTransaction39904060The assistant is allowed to reschedule meetings the company organizes and to share documents already approved for this client. Anything outside those rules would stop here and go to a person.
Update calendarTransaction40604330The meeting moves to the first afternoon slot that works for every attendee.
Retrieve documentsTransaction43304470The latest approved version of the prep package is selected, not an old draft.
Prepare communicationTransaction44704760A short note to the client is drafted from an approved template, with the new time and the package attached.
Complete and confirmTransaction47605200The message is sent, every step is logged, and the person hears a summary of what was done.
The explanation

Speech models now answer in about a second. The value is one spoken request that completes a task across company systems.

What this means

Speaking takes 3.6 of the 5.2 seconds. The value is in the 1.6 seconds after, when the system finds the records, checks permissions and makes the updates. Voice pays off mainly where those connections already exist.

Decision 4 of 6 · how to start

Choose one transaction to run by voice

Pick one transaction where typing or navigating screens slows people down, in field service, sales, operations or customer service. Build it end to end, including the permission check and a handoff to a person.

Suggested owner: Chief Customer Officer
How people use voice today
  1. DictationEmails, documents and AI prompts, spoken. About three times faster than typing.
  2. The foot pedalHold to talk, release to send, with hands kept on the keyboard.
  3. Thinking out loudVoice mode in ChatGPT, Claude or Gemini to prepare and draft on the move.
  4. Hands-busy workTechnicians, clinicians and drivers update records without a screen.
  5. Customer voice agentsService lines that resolve simple requests, with a person on hand.
  6. Requests that get doneOne instruction that updates several systems. Shown below.
PressPulse 2.0 · Aug 2026PractitionerElgato · Jul 2026BenchmarkArtificial Analysis
The research

The leading speech models now start responding within 1.1 to 1.4 seconds and score above 90% on a spoken-reasoning benchmark. That is close to the pace of normal conversation, which makes voice practical for customers, field technicians and sales teams.

Most of the work happens after the person stops talking. The request has to be turned into a structured instruction, matched to the right records, checked against policy, recorded in the system of record and confirmed back to the user. The trace above shows each step and how long it takes.

Customer service leaders increased AI spending by 38% while their overall budgets grew 2%. Customers have one condition: 87% want to be able to reach a person.

The data behind Figure 04

05Change 05 · SoftwareCustom software becomes rational again

Custom software deserves a fresh test against SaaS.

The fact

32%

of organizations have dropped a software purchase because they built the capability themselves

SurveyMcKinsey & Company · Aug 2026
Why it matters

What the software touches sets the risk. Anything that writes to systems of record needs a data and governance foundation first.

Decision 5 of 6

Re-evaluate one differentiated workflow for build versus buy

Owner CIO

Compare building and buying over five years · plus the evidence and how to start

Figure 05

Where should we stop standardizing?

Move a workflow from standard to differentiating and compare its five-year cost.

SurveyMcKinsey & Company · Aug 2026ExperimentMETR · Jul 2025SurveyGoogle Cloud · Sep 2025
AWALI framework. Cost bars in the figure are relative, not measured.
Workflow typeExamplesStrategyWhy
Systems of recordGeneral ledger, payroll, ERP coreBuyEvery company needs the same thing here, and the vendor carries the compliance burden. Buy it and manage it through procurement.
Standard workflowsHR service desk, expense approvalConfigureThe process is much the same across companies. Buy a platform, configure it to fit, and make sure your data stays portable.
Competitive workflowsPricing approval, field dispatch, underwritingBespokeHow you run this process is part of how you compete, and off-the-shelf software forces you to work like everyone else. With AI, building it yourself now makes financial sense.
Core judgmentYour models of risk, demand and customer valueOwnThis is where your advantage comes from. Own the whole system, keep it out of vendors' training data, and invest most heavily in testing.
The explanation

AI has cut the cost of writing software, so building is viable again for the workflows that set you apart. It has not reduced the cost of testing it, maintaining it and owning it over time.

  • −19%measured change in speed for experienced developers using AI, even though they believed they were 20% fasterExperimentMETR · Jul 2025
What this means

The real difference is what the software touches. Tools that sit on top of existing data are low risk to build in-house. Anything that writes to systems of record, touches customers or money, or has to behave the same way across the business needs a proper data and governance foundation first.

Decision 5 of 6 · how to start

Re-evaluate one differentiated workflow for build versus buy

Pick one strategically differentiating workflow that a SaaS product constrains, and compare building and buying on full cost, including testing, maintenance and ownership. Decide which parts your team builds and which foundation layers, such as data and governance, come from a specialist partner or platform. Apply the same review to software renewals above a set value.

Suggested owner: CIO
The research

The research on developer productivity is mixed, and the differences are informative. On new, well-defined tasks, developers were 21% to 55% faster. In METR's 2025 trial, experienced developers working in their own mature codebases were 19% slower, even though they believed they were 20% faster. Google's DORA research links AI use to faster delivery but less stable systems.

The practical conclusion is to build where the logic is specific to your business and the scope is clear, and to fund the review, testing and ownership needed to keep it reliable.

Figure 05b

The productivity gain depends on the task, and developers tend to overestimate it

Reported effect of AI coding tools in controlled trials. Positive values mean faster work or more output. Each study measures something slightly different.

ExperimentGitHub · Sep 2022ExperimentCui · Feb 2025ExperimentGoogle · Oct 2024ExperimentMETR · Jul 2025ExperimentMETR · Feb 2026
Positive values mean faster or more output. METR 2025 participants believed they had been 20% faster.
StudyYearTaskMeasureEffect95% interval (pts)
GitHub Copilot RCT2022One new task (HTTP server)Faster task completion+55%+21 to +89Source
Microsoft, Accenture, Fortune 100 field trials2025Day-to-day workMore completed tasks+26%+6 to +46Source
Google enterprise RCT2024Bounded enterprise taskLess time on task+21%not reportedSource
METR, experienced developers2025Own mature repositoriesChange in speed (time +19%)-19%-39 to -2Source
METR follow-up, returning developers2026Own repositoriesLess time on task+18%-9 to +38Source
METR follow-up, new developers2026Own repositoriesLess time on task+4%-9 to +15Source

The data behind Figure 05

06Change 06 · CreativeCreative becomes a production system

Creative AI is becoming a production system.

The fact

~$3–12

base model cost to generate 30 seconds of video, before editing and review

AWALI calculation · Google & Runway list prices
Why it matters

When versions take days instead of months, the measures change: cost per winning ad, and speed from result to next brief.

Watch area

Baseline creative testing speed

Owner CMO

A 3D landscape built with AI, and how teams use it · plus the evidence and how to start

Figure 06

Higgsfield at a glance

AWALI example · made with Claude, Blender and Higgsfield
Scene
The Teton Range from Driggs, Idaho, in winter, at 5,000 ft
Terrain
Built from USGS elevation data, so the mountains match the real topography
Tools
Claude and Blender built the 3D scene; Higgsfield generated the finished video
Build
A quick build that combined several models in one workflow

Workflow based on a guide by @kyleslyf on YouTube

How teams use it
  1. Write a short shot list and set the look with one key frame from an image model.
  2. Use saved camera presets instead of describing the move in words.
  3. Send wide shots and B-roll to a lower-cost model; keep the best model for the one or two hero shots.
  4. Train a Soul ID if the same presenter or character appears across the campaign.
  5. Generate versions by format and message, then cut, grade and caption in your usual editor.
  6. Test the versions and track the cost of each ad that beats the current one.

Clips run up to 15 seconds per generation, final assembly still happens in an editor, and the capability descriptions are Higgsfield's own.

What it does well
  • Many models, one placeVeo 3.1, Kling 3.0, Seedance 2.0, WAN 2.6 and others sit alongside Higgsfield's own models, so each shot can use the model that suits it.
  • Camera controlDolly, orbit, tracking and crane moves are set when the clip is generated, rather than hoped for in a text prompt.
  • Consistent peopleSoul ID is trained once from 20 or more photos and keeps the same face across every clip and model.
  • From product page to adMarketing Studio reads a product image or link and fills templates for product shots, Meta and Google ads, short UGC-style video and motion graphics.
  • Editing in placeInpainting, relighting, upscaling and layers change a scene without regenerating it from scratch.
Vendor-reportedHiggsfield · Jul 2026Vendor-reportedHiggsfield · Aug 2026PractitionerPractitioner guide · Jun 2026Vendor-reportedHiggsfield · website
The explanation

Platforms such as Higgsfield put the leading video and image models in one workflow, so a small team can test far more creative. Higgsfield adds camera control, consistent characters and ad templates on top of the models. A small team can now produce and test more versions than a traditional production schedule allows.

What this means

When producing a campaign's worth of versions takes days rather than months, the useful measures change: the cost of each ad that beats the current one, and how quickly results feed the next brief.

Watch area · how to start

Baseline creative testing speed

Measure how many ad versions you test each month and how many days it takes to act on results. Then run one bounded test in a single channel with a production platform such as Higgsfield. No enterprise commitment is needed yet.

Suggested owner: CMO
The research

Generated video lists at $0.10 to $0.40 a second, so 30 seconds costs $3 to $12 in model fees before retries, editing and review. Production cost is no longer the main constraint.

Audience reaction still has to be tested rather than assumed. In IAB's survey, 82% of advertising executives believed young consumers feel positive about AI ads; 45% of those consumers do.

07Change 07 · OperationsAI is changing faster than annual plans can respond

44 significant AI developments in nine months.

The fact

11days

median time between major model releases in 2026 (reported)

Market dataOfficeChai · Apr 2026
Why it matters

Annual plans still set strategy. AI operations need a decision point every month.

Watch area

Instrument your AI planning cadence

Owner COO

See how long each AI development waits for a decision · plus the evidence and how to start

Figure 07

How long each AI development waits for a decision, by planning cycle

Choose a planning cycle to see how long each change waits before anyone can act.

Market dataScriptByAI · Sep 2026BenchmarkMETR · May 2026Vendor-reportedOpenAIVendor-reportedOpenAI Help CenterSurveyMcKinsey & Company · Aug 2026
Material AI changes in 2026 to 2 October. Kling 3.0 is dated to February 2026 by Kuaishou; the day shown is approximate. Anthropic's >80% figure is as of May 2026, published June 2026. Model releases are the flagship releases of OpenAI, Anthropic, Google, xAI and Meta plus the leading open-weight releases; dates follow public release trackers, using the first public availability where trackers differ.
DateAreaChange
2026-01-27Model releasesKimi K2.5 (Moonshot AI, open-weight)Source
2026-01-29Coding capabilityMETR: task length doubling every ~4.3 monthsSource
2026-02-05Model releasesClaude Opus 4.6Source
2026-02-15Creative modelsKling 3.0Source
2026-02-17Enterprise tooling & rulesNIST agent standards initiativeSource
2026-02-19Model releasesGemini 3.1 ProSource
2026-02-23Enterprise tooling & rulesgpt-realtime-1.5Source
2026-03-05Model releasesGPT-5.4Source
2026-03-31Creative modelsVeo 3.1 LiteSource
2026-04-07Model releasesClaude Mythos PreviewSource
2026-04-08Model releasesMeta Muse SparkSource
2026-04-15AgentsGartner: 17% have deployed agentsSource
2026-04-16Model releasesClaude Opus 4.7Source
2026-04-22Coding capabilityGoogle: 75% of new code AI-generatedSource
2026-04-23Model releasesGPT-5.5Source
2026-04-24Model releasesDeepSeek V4 (open-weight)Source
2026-04-26Creative modelsSora app discontinuedSource
2026-05-07Enterprise tooling & rulesGPT-Realtime-2Source
2026-05-08Coding capabilityMETR: frontier beyond 16 h, limit of measurementSource
2026-05-19Creative modelsGemini Omni FlashSource
2026-05-28Model releasesClaude Opus 4.8Source
2026-06-09Model releasesClaude Mythos 5 preview and Fable 5Source
2026-06-15Coding capabilityAnthropic: >80% of merged code by ClaudeSource
2026-06-26Model releasesGPT-5.6 SolSource
2026-06-30Model releasesClaude Sonnet 5Source
2026-07-06Enterprise tooling & rulesgpt-realtime-2.1Source
2026-07-08Model releasesGrok 4.5Source
2026-07-16Model releasesKimi K3 (Moonshot AI, open-weight)Source
2026-07-24Model releasesClaude Opus 5Source
2026-07-27Enterprise tooling & rulesEU AI omnibus in forceSource
2026-08-02Model releasesQwen3.8 Max (Alibaba)Source
2026-08-25AgentsMcKinsey: 40% of large firms scaling agentsSource
2026-09-03Model releasesGPT-6 AstraSource
2026-09-06AgentsOpenAI: automated research intern goal metSource
2026-09-07Enterprise tooling & rulesBenchmark index v4.3 resets comparisonsSource
2026-09-10Enterprise tooling & rulesGPT-Live-1 voice APISource
2026-09-21Model releasesGrok 4.7Source
2026-09-22Model releasesClaude Opus 5.5Source
2026-09-22Model releasesGPT-6 Sol and GPT-6 LunaSource
2026-09-22Model releasesMiMo-V2.6-Pro (Xiaomi, open-weight)Source
2026-09-24Creative modelsSora API discontinuedSource
2026-09-28Model releasesClaude Sonnet 5.5Source
2026-09-29Model releasesGPT-6.1 SolSource
2026-09-30Model releasesGemini 4 ArgonSource
The explanation

Under an annual plan, each waited about 204 days for a decision. With a monthly review, the wait drops to 13. AWALI tracked them through September 2026, including 25 major model releases: more change than a once-a-year plan is designed to absorb.

What this means

Annual planning still works for setting strategy, and quarterly cycles still work for allocating capital. AI operations increasingly need a decision point every month.

Watch area · how to start

Instrument your AI planning cadence

Log material AI developments and how long each waits for a decision. Set a trigger: if more than one a month affects your priorities, add a 30-day AI operating review alongside annual strategy and quarterly capital allocation.

Suggested owner: COO
The research

The length of software task AI can complete went from about an hour in early 2025 to about 12 hours by February 2026. Since 2023 it has doubled roughly every 4.3 months. METR notes that its tests can barely measure the leading models, and that the task length at 80% reliability is much shorter.

AI labs report that AI now writes most of their own code: more than 80% of merged code at Anthropic and 75% of new code at Google. Both figures are self-reported and measured differently.

Something that was out of reach when the budget was set can become routine before the end of the year. A monthly review needs a clear trigger for revisiting work when that happens.

Figure 07b

The length of task AI can complete doubles every few months

METR's measure of the length of software task AI can complete with a 50% success rate, by model release date, on a log scale with 95% confidence intervals.

BenchmarkMETR · Jan 2026BenchmarkMETR · May 2026
METR 50% time horizon. Method versions differ between rows; see notes.
ModelReleased50% horizon (min)95% intervalNote
GPT-42023-03-143.51.6–6.9TH1.1METR
GPT-4 11062023-11-063.61.6–7.5TH1.1METR
Claude 3.7 Sonnet2025-02-246032–106TH1.1METR
o32025-04-1612174–201TH1.1METR
Claude Opus 42025-05-2210158–170TH1.1METR
GPT-52025-08-07214117–480TH1.1METR
Claude Sonnet 4.52025-09-2911350–235TH1.0 (Oct 2025 post)METR
Gemini 3 Pro2025-11-18240130–440TH1.1, before 3 Mar 2026 correctionMETR
GPT-5.1-Codex-Max2025-11-1916275–350TH1.0 (Nov 2025 post)METR
Claude Opus 4.52025-11-24320170–729TH1.1METR
Claude Opus 4.62026-02-05719wideCorrected value (~12 h); interval very wideMETR
Gemini 3.1 Pro2026-02-19384240–720Apr 2026 postMETR
GPT-5.4 (xhigh)2026-03-05342180–810Standard method, Apr 2026 postMETR
Claude Mythos Preview2026-04-07≥ 960510–3300Lower bound (≥ 16 h); at the limit of the task suiteMETR
GPT-5.6 Sol2026-06-26678300–2400Jun 2026; cheating attempts counted as failuresMETR

The data behind Figure 07

08Change 08 · GovernanceAI agents need limits built in

Agent risk is about access more than intelligence.

The fact

34%

of organizations apply the same security controls to AI agents as to employees

SurveyOkta · May 2026
Why it matters

The gap between a typical pilot and a governed agent is four controls, far easier to add before launch than after an incident.

Decision 6 of 6

Set real-time controls for AI agents

Owner CISO

Switch an agent’s controls on and off · plus the evidence and how to start

Figure 08

What one AI agent can do on its own, with and without controls

One operations agent with access to seven systems. Choose a setup, or switch controls on and off.

StandardOWASP GenAI Security Project · Dec 2025StandardNIST · Feb 2026
Illustrative scenario: one operations agent with access to Email, CRM, Finance, Documents, Calendar, External tools, Customer comms. Nothing is executed.
ControlWhen onWhen removed
IdentityEvery action is signed by the agent's own identity, with a named owner.Actions run on a shared login, so nobody can tell which agent did what.
Permission scopeThe agent can reach only the systems and actions this job requires.The agent has broad access, so deleting records, wiring money and calling any API all become possible.
Approval thresholdsExternal emails, file shares and customer messages need a person's approval.Actions that can't be undone go ahead without anyone approving them.
Spending limitsPayments are capped at $1,000, and the agent can't make purchases or issue refunds.There is no limit on how much money the agent can move.
Rate limitsThe agent can take at most 20 actions a minute, so mistakes stay small.The agent can take thousands of actions a minute, so one bad instruction can repeat many times before anyone notices.
Audit loggingEvery action is recorded with who did it, what went in and what happened.Nothing is recorded, so incidents can't be reconstructed.
Human escalationUnclear or out-of-policy cases go to a named person.The agent decides for itself how to handle exceptions.
Kill switchAn operator can stop the agent and revoke its access in the middle of a task.Once a task is running, it can't be stopped.
The explanation

Enterprise exposure is determined less by an agent's raw capability than by the authority, access and controls surrounding it. Agents can now send email, move money and change records on their own. A written policy is unlikely to stop one in the middle of a task; controls built into the systems it uses can.

What this means

Only 34% of organizations apply the same controls to AI agents as to employees. The gap between a typical pilot and a governed agent is four controls, and each is far easier to put in place before launch than after an incident.

Decision 6 of 6 · how to start

Set real-time controls for AI agents

Make an inventory of every AI agent and automated account. Give each one a named owner, its own identity, spending and activity limits, approval thresholds, logging, an escalation path and a tested way to shut it down.

Suggested owner: CISO
The research

The necessary controls are well understood: a separate identity for each agent, only the access it needs, approval thresholds, logging, ongoing testing and the ability to revoke access immediately. Recent incidents show why they need to be enforced while systems run rather than written into policy. In 2025 a coding agent deleted a live production database during a code freeze, and AI carried out 80% to 90% of the work in a state-linked hacking campaign. In 2026, evaluation agents reached a third party's production systems.

Practice is behind. Only 34% of organizations apply the same controls to AI agents as to employees. OWASP's first list of top risks for agentic applications includes identity abuse, misuse of tools and agents acting outside their instructions, and NIST has started standards work on agent identity and authorization.

The data behind Figure 08

Your 90-day operating agenda

You now have the eight changes. Here are the decisions to make.

Each of the six decisions has a suggested owner. In an AWALI workshop, we turn them into named owners, dates and a first operating review within 30 days.

Discuss your challenge

About this briefing

One author · 65 sources · 44 AI developments · 8 changes · 6 decisions

Conceived, researched, written and built by Scott ThomasAWALI · Current to 2 October 2026

    Next briefing

    The next one, the day it is published.

    One email per quarter, when a briefing is published. Nothing else.

    Used for the briefing and nothing else. Privacy

    Sources and method · 65 sources

    Where every figure comes from

    • Figures are reported as published by each source as of the date shown. This edition was compiled on 2 October 2026, and model and pricing data reflect that date.
    • Where a shorter summary would overstate a finding, we use the source's original, narrower wording. AWALI has not independently reproduced these studies.
    • Self-reported figures, press reports, forecasts and AWALI calculations are labeled wherever they appear.
    • Every chart is drawn from the values shown in its table view. Figures we could not confirm were left out rather than estimated.
    • The interactive scenarios (the routing policy, the decision simulator, the voice trace, the build-versus-buy comparison, the ad-testing example and the agent controls map) illustrate how these systems behave. They are not measurements.

    Independent benchmark 6

    Independent research 5

    RCT / experiment 5

    Survey 8

    Market data 4

    Published pricing 6

    Vendor-reported 12

    Practitioner guidance 2

    Incident 3

    Regulation and standards 3

    AWALI illustrative scenario 1

    Prior AWALI briefing 2

    AWALI platform 1