Autonomy in AI is accelerating, but are the controls in place to trust it? ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­    ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­  
View in browser
AI_Pulse_Newsletter_Logo_2025_208px-w

Hello from ABBYY's AI Team

Welcome to the October edition,

 

Handing AI the keys is easy. Checking the flux capacitor first is the part everyone skips. The pace at which we’re granting AI agents autonomy is staggering, yet the controls and visibility needed to trust these systems lag behind.

 

First up, Diogo Almeida, co-inventor of RLHF and InstructGPT, explains why LLMs should stay away from high-stakes decisions without human oversight. His new venture, Jev, returns decisions with an inspectable confidence estimate that software can act on. Next, BCG’s Applied AI Index 2026 shows that while 42% of companies plan to grant AI agents autonomy by 2030, only 5% have the critical controls in place today. Finally, one in four AI agents is running unmonitored, and bolting on yet another tool won’t fix that. Unified observability will.

 

The thread running through all three: calibrated confidence, governance and reliable data are what make autonomy possible. Skip them, and you’re hitting 88 mph with no idea where you’ll land. Dig in to read the ABBYY expert’s take on each of these! 

 

Max Vermeir, VP of AI Strategy

AI_Pulse_Newsletter_Dividers_1200x440-blue2-min-cropped

AI News You Need to Know: Expert Take

jon-knisley_ascend-2026

Jonathan Knisley, Global Process AI Lead, ABBYY

AI adoption is obscuring system visibility

One in four AI agents run unmonitored within enterprise environments, which leads to operational risk as the technology autonomously alters code, configurations and infrastructure. It’s fueling a surge in observability adoption, with eight in 10 organizations planning to deploy AI application observability within the next 12 months.

AI workloads are driving demand beyond traditional observability tools, which are designed to measure the performance of cloud estates and other core systems. The shift is contributing to observability tool sprawl. Unified platforms that absorb AI capabilities rather than adding new tools will allow enterprises to capture the benefits of AI without adding additional sprawl and complexity to the operational environment. 

Learn more  >

Slavena-Hristova_ascend-2026

Slavena Hristova, Director of Product Marketing — Document AI, ABBYY

Before you give AI the keys

BCG's Applied AI Index 2026 offers an encouraging message: AI leadership isn't reserved for tech giants or digital natives. Any company can reinvent itself and thrive.

 

BCG's formula for success combines strategic clarity with applied AI: rethinking operating models, transforming data, and reinventing how people and AI work together. And measuring success where it matters, on the P&L, rather than through impressive-looking AI dashboards.

 

What struck me most, however, is how seriously, or rather how unseriously, companies seem to be taking the question of control.

 

While 42% of companies expect to grant AI agents autonomy by 2030, only 5% have all the critical controls in place today.

 

If we want AI agents to take on greater responsibility, we must first give them the guardrails to operate reliably. What makes autonomy possible is governance, trustworthy data, validation, and accountability. They should not be viewed as obstacles.

Learn more  >

Marlene-Wolfgruber_ascend-2026

Dr. Marlene Wolfgruber, Product Marketing Lead AI, ABBYY

The LLM Pioneer Who Says: Keep LLMs Away From Decisions With Stakes

When someone who helped build ChatGPT starts a new company, the question is what he builds. Diogo Almeida, who co-invented RLHF and InstructGPT, sums up his talk “AI: Too Good to Be True, Too Bad to Be Useful” bluntly: today’s LLMs are optimized for text and assistance, so keep them away from decisions with stakes and keep humans in the loop. His answer is Jev, a model for machines: a typed decision with an inspectable confidence estimate that software acts on like an API call.

 

We read it as a signal, when even the model builders now recognize that reliable automation needs calibrated confidence, typed outputs and intelligent human escalation. TypeSafe calls it a new primitive. At ABBYY we have been working this way ever since, tying every value to its source page. Reliability starts there.

Learn more  >

Fun Facts

 

Open-weight models ran the majority (56%) of Vercel's gateway tokens for the first time in August, up from 7% in December. 

 

Source: Vercel >

AI by the Numbers

data-reliability-banner-v4_ai-pulse_numbers

Source: Informatica >

ABBYY in the News

  1. The Register highlights ABBYY’s FineParser, explaining how its CPU-based OCR preserves document structure, tables and layout before passing structured data to LLMs. The article positions traditional OCR as a useful, deterministic foundation for modern AI workflows, rather than something AI has made obsolete >
  2. IT-SICHERHEIT covers ABBYY’s State of Intelligent Automation 2026 research focusing on the growing gap between AI adoption and effective governance. It highlights ABBYY findings including 54% of leaders saying AI adoption is outpacing governance, unclear accountability for AI outcomes, and gaps in incident response and technical controls. Roman Kilun is quoted on the importance of clear responsibility, controls and reliable data for safe AI adoption >
  3. InformationWeek features insights from Jon Knisley, who argues that legacy systems are not an obstacle to AI success but a competitive advantage. He explains that decades of transaction history, business rules, and institutional knowledge can become an organization's "AI moat" when paired with purpose-built AI and strong data foundations, enabling faster, more practical AI adoption without costly rip-and-replace projects >
AI_Pulse_Newsletter_Dividers_1200x440-gray-min-cropped

ABBYY in Action: Innovation Spotlight

state-intelligent-automation-report-sept-2026_ai-pulse-newsletter-square-2026

State of Intelligent Automation Report: Who Answers for AI? The Governance Gap

ABBYY’s most recent survey, conducted in July 2026, captured the perspectives of 1,200 IT leaders and decision makers from 20 industries. It explores who organizations believe should answer for AI failures, the impact of governance on AI success, the barriers preventing stronger ROI, and how businesses are strengthening governance frameworks to support responsible and scalable AI adoption. 

Access the report
AI-pulse-podcast-logo-2025

The Soul Factor: Why AI Adoption Is Really About People

Why do so many AI initiatives fall short, even when organizations invest heavily in them? Technology strategist and Soul vs. Scale author Dado Van Peteghem joins the AI Pulse Podcast to explore the often-overlooked human side of AI transformation. Tune in for a valuable framework and a reality check for turning AI ambition into lasting impact.

Listen now

AI in Practice: Real-World Applications & Case Studies

customer-story-background

Hybrid Implementation Outperforms Homegrown LLM-First Approach

A global fund administrator explored building its own document processing capabilities based on generative AI models, but ultimately choose ABBYY’s purpose-built Document AI approach to ensure consistency, reliability, and accuracy in its financial workflows. The result: 96% straight through processing. 

Read the story

Quick Resources & Events

servicenow_ai-pulse-newsletter-square-2026 (1)

How ABBYY and ServiceNow Turn Documents into Action

ABBYY transforms documents into structured, business-ready information. ServiceNow transforms that information into business action. Together, we create an intelligent automation pipeline where documents no longer pause the business - they move it forward.

 

Read more on the blog  >

abbyy-ascend-singapore-2026_ai-pulse-newsletter-square-2026

Catch Up on Ascend Singapore

Missed Ascend Singapore? Explore the presentations and content from the event, covering the latest in Document AI, Gen AI, AI governance and practical examples of how organizations are putting AI to work across business processes.  

 

Access the content here  >

ciosignature_ai-pulse-newsletter-square-2026

CIOSignature - BFSI New York

Join us in New York on 10 November 2026 for an exclusive, interactive roundtable bringing together senior BFSI leaders to explore how enterprises are using Intelligent Document Processing (IDP) to automate document-driven work. Discover practical strategies to identify high-value AI opportunities and achieve measurable ROI as you scale AI and automation initiatives.  

 

Register Now  >

becker-hospital-review_ai-pulse-newsletter-square-2026

Captured but not posted: How revenue cycle teams close the loop on correspondence 

Join Naviant and ABBYY on Thursday, October 22, 1:00 PM CDT, for a Becker’s Healthcare panel designed for healthcare providers on automating revenue cycle correspondence. Learn how remittance advice, EOBs, denial letters, and refund requests can move from a scanned lockbox batch to the correct patient account with minimal manual effort, while your staff stay in control of exceptions that require judgment. 

 

Register now  >

Facebook X LinkedIn YouTube ABBYY Blog

© 2026 ABBYY Solutions Limited 

You have received this e-mail because you are an ABBYY customer or partner, have requested further information on our products, solutions or services, registered for an ABBYY trial version, and/or agreed to receive news from ABBYY. 

This e-mail is sent to you by: 

ABBYY Solutions Limited, Legal address: 25 Aphrodite Street, Office 204 (2nd floor), 1060, Nicosia, Cyprus

www.ABBYY.com/company/contact-us 

Director of ABBYY Solutions: Anna Terzi - Registered at Registrar of Companies, Nicosia, Cyprus HE 131582 

Vat Reg. No.: CY10131582D 

This email was sent to  info@abbyy.com. 

Privacy Policy 

You may unsubscribe from future emails at any time. Should you have any questions don't hesitate to contact us.

Unsubscribe Manage preferences