Things continue to accelerate in the Conversational AI market.

Omilia is raising $67M in a Series B. This comes six years after its Series A in May 2020. During that period, its live ARR has grown more than 10x to over $60M, representing roughly 45% CAGR.

The long-standing player with deep roots in voice solutions has operated somewhat under the radar, despite being featured in the latest Forrester Wave, Gartner Magic Quadrant, and IDC MarketScape for Conversational AI.

Founded in 2002 as a Nuance value-added reseller, Omilia pivoted to a product business in 2008 and has since implemented several large-scale deployments at Capital One, Discover, RBC, Taco Bell, DWP, and PSEG.

The funding also comes with two notable hires: Nick Delis as Chief Revenue Officer and Ryan Kam as Chief Marketing Officer. That signals an ambition to step out of the shadows and compete more aggressively in a market that is heating up quickly.

Interesting development in the conversational AI space.

Yellow.ai announced it is going public via a SPAC merger with Bluerock Acquisition Corp.

As part of the process, it provided several interesting data points: revenue of $34M and enterprise value of $300M.

The most notable aspect is the company’s strategy.

Yellow.ai sees the BPO market as a major growth opportunity. The company plans to use the $200M raised through the transaction to acquire and roll up BPOs, transforming them into AI-native operations powered by its platform.

That ambition mirrors the direction companies like Crescendo have started to pursue.

At the same time, incumbent BPOs are not standing still and investing aggressively in building their own technology platforms. In its latest quarterly results, TP highlighted the growing traction of its TP.ai platform, with 1,100 AI projects now in production and 270 launched in the last quarter.

With Yellow.ai becoming public, we’ll gain much greater visibility into the transformation of the BPO market.

The investor presentation is worth a read: here

Earlier this month, Zoom made a bold move by acquiring sales intelligence specialist Common Room.

The acquisition surprised many, myself included, even though I was aware of Zoom’s growing ambitions in sales and revenue technology.

Common Room was founded in 2020 as a community intelligence platform. It raised more than $50M across three funding rounds and has approximately 180 employees, suggesting annual recurring revenue around $20M.

The company captures buying and engagement signals across public and community ecosystems, including GitHub, Reddit, Discord, Slack communities, and, of course, LinkedIn. While it initially focused on helping companies uncover the “dark funnel,” Common Room evolved into a broader sales intelligence platform centered on a CDP-like data foundation. It unifies first-, second-, and third-party data into what it calls a customer journey graph, with capabilities including identity resolution, waterfall enrichment, and signal aggregation.

Common Room allows users to layer workflow logic or AI agents on top of its graph to score both individuals and companies, and orchestrate targeted sales plays. The resulting intelligence can be consumed through Common Room’s own seller interface, pushed into engagement platforms, or exposed to AI agents through MCP.

Zoom formally entered the SalesTech market in April 2022 with the launch of Zoom IQ for Sales, its conversation intelligence solution. I say “formally” because Zoom already had footholds in sales workflows through Zoom Meetings and Zoom Contact Center. In September 2023, the company rebranded Zoom IQ for Sales as Zoom Revenue Accelerator, later expanding it into an umbrella for its sales solution, adding capabilities such as deal intelligence, coaching, next-best actions, and automated CRM updates.

Zoom has architected its sales offering on top of a comprehensive communications platform. To understand the full breadth of its portfolio, you need to look beyond Zoom Revenue Accelerator and include platform capabilities such as Zoom Contact Center with its built-in auto dialer, Zoom Virtual Agent, and more.

Combined, the broader Zoom platform, Zoom Revenue Accelerator, and Common Room provide strong coverage across the revenue stack. Using my new Revenue Blueprint, I mapped the full breadth of the Zoom portfolio’s sales capabilities.

Zoom is now a company to watch in the SalesTech market, across both B2B and B2C. The Common Room acquisition adds a solid sales intelligence layer to its broader communications and revenue platform and propels it into the group of SalesTech consolidators.

Zoom has the ingredients to become a meaningful player in the evolving revenue technology landscape. Capturing this opportunity will require it to bring together its portfolio of capabilities, deepen its functionality across key SalesTech categories, and expand its go-to-market motion to engage a broader set of buyers beyond its traditional communications and IT audience.

Verint held its annual customer and partner conference, Engage, in June.

The company was taken private in August last year and later combined with Calabrio. The merger closed in February, and Engage, held four months later, provided the first opportunity for the new Verint to unveil its refreshed visual identity and outline its direction.

Dave Rhodes addressed concerns from its two large workforce management (WFM) customer bases upfront, reassuring them that both Calabrio WFM and Verint WFM will continue to be supported, with no forced migrations.

At the same time, Verint is integrating Calabrio’s best-of-breed technology components into its CX automation platform.

The central message from the new Verint is its renewed commitment to an outcome-driven go-to-market (GTM) strategy. Verint continues to focus conversations on the business outcomes its 50 bots can deliver, rather than on technology or agentic capabilities. Its approach is incremental and allows enterprises to layer bots onto any existing contact center infrastructure. At a time when agentic narratives dominate, Verint is choosing to zig toward business outcomes while the market zags toward technology narratives.

To support this outcome-driven strategy, the company introduced a risk-free trial last year: if ROI is not demonstrated within the first 90 days, customers pay nothing. ROI calculators are embedded directly into the platform. Verint has also introduced workshops to demonstrate the art of the possible with its bots and identify priority use cases.

BT showcased a compelling example of this outcome-driven approach at the event. Anthony Cass detailed a phased deployment that scaled from 450 to 4,500 sales agents. Using three bots, Coaching, Wrap-up, and CX Scoring, BT reported improvements in real-time agent guidance, cross-sell and upsell, onboarding, and churn reduction. Cross-selling alone drove a $9M increase in annual revenue.

To reinforce this strategy, Verint is making substantial investments in GTM enablement and methodology. The goal is to equip teams to shift customer conversations from technology discussions toward measurable business outcomes.

On the product front, Verint Intelligent Virtual Assistant (IVA) deserves mention, as the company’s self-service offering is often overshadowed by its broader automation portfolio. It has now surpassed the 100-customer mark.

Among the product announcements, two stood out.

The first is Agent Factory, a control layer for its bots that enables customization and orchestration. It allows enterprises to orchestrate workflows across bots and humans, for example, combining post-interaction summarization with a proactive follow-up when customer sentiment indicates significant frustration.

Verint also unveiled a Desktop Intelligence product that uses AI to analyze screen captures, understanding both conversation outcomes and what actually happened during the interaction. This new layer of insight is a meaningful step toward unlocking the value of more than 20 years of interaction data. The solution can also be used to validate that what an agent says is aligned with what is entered into back-end systems.

After a somewhat quiet period, Verint reminded the market that it is a force to be reckoned with. The breadth of its automation portfolio, combined with a differentiated outcome-based go-to-market, gives Verint a distinctive position in a market increasingly centered around agentic AI.

Founded seven years ago, Warmly, has reached $7.5M ARR with 400 customers. It has seen accelerating penetration within the HubSpot ecosystem, which has grown to represent more than half of its customer base in just two years.

At the core of Warmly, is website visitor intelligence, a capability now claimed by many vendors as a checkbox feature, but with widely varying degrees of accuracy and coverage.

Warmly, has built a strong reputation here, claiming ~60% company identification and ~45% person-level identification. But HubSpot already had a strong foundation in this area through its Clearbit acquisition.

The differentiation in Warmly, is not the visitor identification, but what is done with it.

Where Warmly, shines is its ability to operationalize these signals into real-time GTM workflows. After intent scoring, it can trigger actions across the revenue stack, including Slack alerts, CRM updates, lead routing, sales task creation, AI chat engagement, and ad audience activation.

Again, HubSpot already offers a Customer Agent that can be configured to qualify inbound leads. In its most recent earnings call, the company actually reported 2x adoption growth, reaching 2,500 customers and roughly doubling usage. However, this agent is primarily oriented toward customer service and support scenarios.

Warmly,’s approach is more purpose-built and modular. A TAM Agent identifies and enriches new prospects and records them in systems of record, a Signal Agent monitors website engagement and buying intent, and an Inbound Agent captures and converts visitors in real time by triggering the appropriate workflows, with all three working in concert. The result is not just enriched visitor data, but coordinated GTM execution across sales and marketing systems.

This acquisition mirrors the one of Qualified by Salesforce: the same core use case around inbound engagement and demand capture, and a similar approach based on specialized, purpose-built agents.

It also reflects HubSpot’s continued move towards building an integrated revenue stack.

Genesys announced it is acquiring Pinkfish.

Pinkfish is a relatively new, 30-month-old workflow orchestration platform. Despite its founders’ roots in customer service and contact centers, Pinkfish has so far targeted enterprise workflow automation broadly rather than CX specifically.

It can be viewed as a modern, AI-native evolution of automation platforms such as UiPath or ServiceNow, with a strong emphasis on simplifying and democratizing process automation through natural language and AI. Workflows are built from nodes/steps that can be deterministic, agentic, or involve tool calls. The company lists 500 integrations and 25,000 MCP tools.

Pinkfish applies agentic capabilities primarily at the step level to create workflows or agents using natural language, dynamically discover and select tools, and execute simple tasks. This enables a hybrid model where deterministic workflows can leverage AI-driven step execution. The ability to define workflows or steps through natural language is making process automation more accessible to business users.

Genesys has been aggressively expanding its agentic capabilities. It introduced its Agentic Virtual Agent earlier this year, featuring Large Action Models (LAMs) via its partnership with Scaled Cognition alongside traditional LLM-based approaches. LAMs aim to reduce hallucinations, in particular through curated training datasets, including synthetic data and simulated actions, which improve the accuracy of tool use and action execution. Pinkfish extends these capabilities by enabling self-service requests to be executed and fulfilled across enterprise systems.

Perhaps more significantly, this acquisition marks a meaningful step into back-office customer interaction fulfillment and end-to-end process execution. Genesys already had Work Item and Case Management capabilities, and the press announcement explicitly links the acquisition to the latter for long-running processes. The signal is clear: Genesys isn’t just orchestrating customer experiences anymore. It wants to own a larger share of CX workflow execution.

AI-related layoffs and workforce reductions continue to dominate headlines, with contact center employment often cited as being most exposed to AI-driven disruption.

Indeed, Agentic AI is reaching a level of capability where it can absorb a large share of first-level and transactional service interactions. Recent research highlights the scale of potential exposure: Anthropic found customer service roles to be among the most exposed to AI, with up to 70% of tasks potentially handled by AI. Similarly, Goldman Sachs estimated that a significant share of customer service roles are exposed to AI-driven automation.

The US Bureau of Labor Statistics (BLS) recently published its annual employment-by-occupation report, offering a timely opportunity to assess the state of the contact center and customer service workforce.

For the second year in a row, employment of Customer Service Representatives has declined by 5%. The label Customer Service Representatives is somewhat misleading, as it encompasses only a fraction of customer service roles. Another category, Computer Support Specialists, points in the opposite direction. Last year, employment grew 2% after two years of 1-2% decline. This was a surprise, given that customer support automation is one of the most mature use cases for AI.

Combining the two categories shows a 7.5% contraction over the past three years. While Gartner’s 2023 prediction that AI will drive a 20-30% reduction in customer service and support agents by 2026 has not materialized, these numbers signal a structural decline.

But they only capture a fraction — 75% and declining — of the broader customer-facing workforce using customer communication technologies across the customer lifecycle. They do not capture the expansion of customer engagement activities, both in volume and across a broader set of roles across the enterprise.

First, we often assume a stable workload of interactions. In reality, while AI is absorbing a substantial share of interactions, volume and complexity continue to surge, offsetting much of the efficiency gains from automation:

• The scope of activities handled virtually keeps expanding; COVID demonstrated that there is effectively no ceiling to what can be done remotely.

• Processes are becoming more complex, driven in particular by mounting regulatory and compliance requirements.

• Digitization is spreading across more processes, which generate exceptions, adding to customer service workload. In its Inner Circle Guide to Customer Interaction Analytics, ContactBabel found that 80% of complaints reaching contact centers are caused by back-office failures.

• In parallel, the contact center’s managed communication infrastructure has extended its scope beyond traditional service into adjacent functions, including sales and service delivery, such as financial advisory or healthcare interactions.

• Finally, the looming rise of agents acting on behalf of consumers will drive interaction volumes to new highs.

In its AI Automation vs. Human Effort internal study, Zendesk found that automation was actually driving a significant increase in interaction volumes, with larger increases occurring at higher levels of automation. The study reported up to a 200% increase in volume at a 60% automation rate, suggesting that Jevons Paradox is at play in customer service.

Second, the occupational definitions we rely on are increasingly misaligned with how customer-facing work is structured:

• The contact center represents a relatively recent addition to customer service. Its structure reflects a traditional pattern of decomposing customer service work across the enterprise, mirroring Taylorism in early 20th-century manufacturing.

• It is a carved-out “front-end layer” for customer service, optimized for rapid responses through large pools of agents handling high-frequency inquiries, while other tasks are offloaded to specialized departments. Over time, these functions have become even more specialized, creating fragmented customer service experiences, longer resolution times, and friction across organizational boundaries.

• AI is not only becoming the “front door” to service, delivering instant responses, it is also assisting customer-facing associates with specialized tasks, enabling a reaggregation of jobs. Instead of funneling customer communications through the contact center to dispatch the work, a larger pool of the workforce will interact directly with customers.

Taken together, these dynamics suggest the contact center, as a distinct organizational construct, is likely to dissolve into a broader customer-facing operating model.

Rather than framing this shift as AI-driven job elimination, the focus should be on how customer engagement work is redistributed between AI and humans across the enterprise, and how that redistribution reshapes customer-facing roles.

A corollary of this transition is that many entry-level customer service roles that have traditionally served as an entry point into the workforce will disappear, and organizations will need to rethink pathways for employees entering the workforce.

NiCE held its annual conference in Orlando earlier this month.

The event was rebranded as NiCE World, reflecting both the company brand identity introduced a year ago and its broader ambition in the CX space. The event was structured around a show floor designed to facilitate interactions among the 3,000 customers, partners, and employees in attendance.

Under Scott Russell, NiCE has re-centered its GTM around partners, especially global systems integrators (GSIs). Partners now drive ~70% of net-new enterprise logos, with GSIs contributing 14% of ACV, roughly 3x their contribution a year ago. In international markets, close to 90% of the company’s business is partner-led via resale or co-sell.

Scott Russell reasserted the company’s position in the CX AI space with his characteristic conviction, stressing that AI is built in, not bolt on, and emphasizing the importance of a unified platform and its ability to scale.

NiCE reaffirmed Cognigy’s contact center neutrality. NiCE’s strategy is to offer multiple entry points into its suite, CCaaS (contact center), workforce management, or Cognigy’s conversational AI platform.

Nine months after the Cognigy acquisition closed, the integration is effectively complete. Cognigy now has access to NiCE’s proprietary datasets, including vast volumes of historical conversational data. These datasets include highly valuable human-handled interactions, typically the most complex and business-critical conversations, providing Cognigy with a strategic differentiator over other pure-play conversational AI providers. Cognigy has also become the agentic and conversational AI layer of CXone, bringing its best-of-breed capabilities into the NiCE platform.

The event showcased learnings from large-scale AI deployments. One point that caught my attention was that several customers explained they chose to tackle the most complex problems first with AI, rather than starting with simpler use cases, as has traditionally been the adoption pattern. This suggests new levels of confidence in the technology.

Philipp Heltewig offered a compelling example of Cognigy handling over 80% of the 20 million customer conversations with Lufthansa during a one-week strike, including cancellations and flight re-arrangements. He also warned of the imminent rise of consumer-facing conversational agents, urging enterprises to prepare for a significant increase in AI-driven traffic. These examples illustrate the technology’s progress, including instances of AI outperforming humans or managing situations humans could not.

Finally, Jeff Comstock and the product team provided a private preview of several new products and capabilities worth discussing:

• Agentic Engagement Plane is a new layer that allows AI and human agents to operate as a unified workforce. It uses the concept of engagement paths to define the resources that can handle conversations across NiCE and third-party conversational AI and CCaaS platforms. It builds on a global view of the workforce and supports diverse human-in-the-loop models, including full takeover, silent assist, and mid-flight takeover.

• It is supported by the next evolution of Experience Memory, which maintains customer identification and authentication, conversation context, and journey history, including reasoning and decision traces.

• Experience Optimization is evolving into Agentic Analytics spanning conversation, workflow, and outcome data. It leverages a set of domain-specific AI agents to analyze this dataset to identify new automation opportunities and improve existing ones. It draws on a new Screen Intelligence capability to understand everything that happened.

• NiCE is working on a redesigned agent workspace to weave AI agents into the conversation stream, whether handling the conversation, providing recommendations, or requiring human validation.

• AgentForge is a new capability that generates AI agents from artifacts such as documents or specifications.

• Guardian AI is a new governance and control layer that monitors both human and AI agents to enforce guardrails and policies and surface risks.

These building blocks demonstrate NiCE’s continued advancement toward managing customer conversations and resolution workflows across the full span of customer engagement through a unified human and AI workforce.

Outreach held its annual user conference, Unleash, in early June

This year’s edition was a more intimate event, enabling deeper interactions with and among its larger customers and partners while reflecting the company’s steady move upmarket. It provided the perfect opportunity to take stock of the platform it has put together.

There are still parts of the market that do not yet fully appreciate the breadth of sales jobs to be done that it supports, from prospecting and deal management to account development, retention, coaching, and forecasting.

Here is a mapping of its suite’s capabilities to my new Revenue Blueprint:

Speakers from Cohesity, Databricks, Freshworks, OpenAI, SAP, ServiceNow, and SolarWinds shared their go-to-market (GTM) transformation journeys. Two things stood out: the scale of some of these deployments and the central role Outreach has played in enabling these transformations.

The foundation of the Outreach stack is a revenue data platform. The company was among the first vendors to expand its deep CRM integrations with two-way integrations to enterprise data warehouses. It consolidates engagement activity – emails, calls, meetings – along with first-party data from CRM, data warehouse, and website activity, third-party data and signals via its Smart Data Enrichment capability and Research Agent, as well as knowledge assets such as collateral and messaging guidelines.

The federated context can be used by its own agents or made available to third-party agentic platforms via its MCP server. At the event, it detailed its two-sided partnership with OpenAI, making its data, context, and agents available natively within ChatGPT and Codex.

Today, Outreach boasts one of the richest integration ecosystems across CRM systems, data warehouses, and LLM platforms:

CEO Abhijit Mitra shared his conviction that architecture matters and outlined how its Agentic AI Platform for Revenue Teams turns myriad signals into coordinated action.

Outreach has embraced agentic AI, with a vision of agents as teammates that get things done. It has created a suite of specialized agents, notably its Revenue Agent, which builds on segmentation by layering enrichment and signals to assemble prioritized targets to pursue, and its Deal Agent, which keeps an accurate view of opportunities through automated activity capture, stage progression, summaries, and insights. Earlier this year, it added an Agent Studio to its agent portfolio, enabling customers to build their own. At the event, it announced both an agent marketplace and several new purpose-built agents.

By offering a broad set of focused agents, Outreach aims to accelerate agentic adoption across its customer base. It also introduced an AI Maturity Model for sales, enabling revenue organizations to assess their operations and prioritize their agentic progression.

Outreach serves sellers, managers (leaders), and RevOps via unified workspaces where they land on a “home” page that provides a global view of their book of business and prioritized actions. At the event, the company previewed expanded AI recommendations with new capabilities such as Deal Strategy for sellers or Win Paths, helping leaders identify the best options to reach their targets. The company is also making its capabilities available to users wherever they are, via Slack or its mobile application.

The event also provided an opportunity for Outreach to showcase Omni, its newly introduced conversational agent. Omni a natural language interface built on top of its data platform that leverages revenue context to allow users to ask business questions and receive both answers and recommendations.

Outreach is at an inflection point. The platform has come a long way since its debut as a sales engagement solution and now spans a much broader set of sales and revenue workflows.

Yet, the company does not enjoy its fair share of mind in the consolidating sales and revenue technology landscape. Combined with the lingering perception of being primarily an inside sales prospecting tool, this places the company in a position where it has to continuously articulate the full scope of its platform, which can make the overall value proposition somewhat overwhelming.

It is time for Outreach to streamline its positioning and tackle its brand perception more directly, reframing itself as an all-in-one platform for B2B sales.

Meta held its 2026 Conversations messaging conference in London earlier this month and made several significant announcements about its business messaging portfolio.

US-based observers often underestimate how pervasive WhatsApp has become, not just for consumers but for business engagement across the globe. WhatsApp's reach extends well beyond consumers: more than 1 million businesses use Meta's messaging and automation capabilities across WhatsApp and Messenger. In Brazil, WhatsApp is the primary channel consumers use for business communication, a pattern also reflected elsewhere: Air France exchanged 33 million messages with passengers over 12 months, making WhatsApp its #2 digital contact channel.

So far, Meta has been offering its Meta Business Platform directly to enterprises and through partners for small and midsize businesses. Meta Business Platform provides APIs to manage conversations, enabling use cases such as chatbots, marketing campaigns, and appointment scheduling.

At its London event, Meta made four announcements:

• Meta Business Agent is a ready-to-use customer engagement agent for SMBs, embedded across WhatsApp, Messenger, and Instagram, that handles conversations, including answering customer inquiries, qualifying leads, booking, taking payments, and escalating to humans.

• Meta AI Agent Platform targets large enterprises looking to build similar customer engagement use cases across Meta's three messaging channels, with the ability to apply customizations and connect to existing systems.

• Meta Business Agent Home is an application that lets businesses manage their messaging conversations at scale.

• Enhancements to WhatsApp Discovery that enable customers to find businesses directly inside WhatsApp.

While messaging remains promising, it has not yet been adopted as a broad-based business channel and has largely been confined to specific use cases. What is missing is enterprise support for a wider set of end-to-end sales and service experiences for these touchpoints to reach their potential. Meta has the market footprint to drive this shift. When it launched its Business AI program last March, I questioned whether it could finally crack enterprise sales and service. It increasingly looks like it can.

Earlier this month, I published a new revenue technology blueprint. I now want to start a series that takes a closer look at its key building blocks, starting with systems of record and the workflow and agentic layer.

Organizations run on systems of record. They capture data, embed business logic, and serve as the operational backbone of the business. It all started with CRM, which became the system of record for commercial relationships and pipeline management: accounts, contacts, opportunities, and deals.

The rigid CRM data model, centered on the lead-to-account-contact-opportunity lifecycle, quickly became a constraint for marketers. This gave rise to Marketing Automation Platforms (MAPs) such as Eloqua and Marketo. Built around a flatter, person-centric model, they became the systems of record for first-party lead data and campaign engagement.

The same rigidity later drove the emergence of Account-Based (ABX) platforms from vendors such as 6sense and Demandbase. They became the systems of record for account-level intent, signals, and engagement.

Similar limitations for post-sales motions led to the emergence of Customer Success Management (CSM) platforms from vendors such as Gainsight and Totango. They aggregate product usage and other account-level signals, becoming the systems of record for customer health and expansion by combining product, relationship, and commercial data.

The rise of product-led growth also spawned a new, albeit short-lived, category of Product-Led Sales (PLS) platforms from the likes of MadKudu and Pocus. They became systems of record for user-level product engagement, combining telemetry, user and account context, and third-party enrichment to support freemium-driven sales motions.

This proliferation of systems drove enterprises to build Data Warehouses (DWH) on platforms such as Databricks and Snowflake to create a unified view of their customers. These data warehouses became the systems of record for enterprise reporting and AI modeling, consolidating customer, account, and event data across the business.

Their analytics-oriented nature made them ill-suited for real-time marketing and sales engagement. This gap led to the emergence of Customer Data Platforms (CDPs) from providers such as Segment and Tealium, which became the systems of record for unified first-party events and signals, customer profiles, third-party attributes, and identity resolution.

More recently, two new platform categories have emerged. Customer Engagement Platforms (CEPs), such as Braze and Klaviyo, combine CDP capabilities with real-time customer engagement across channels, becoming systems of record for customer interactions and engagement orchestration. At the same time, Revenue Orchestration Platforms (ROPs), such as Gong and Outreach, have consolidated conversations, sales activities, and customer interactions, becoming systems of record for sales engagement and execution.

Today, most businesses rely on a combination of these systems. Taking inventory of your data estate is a critical first step in designing an integrated revenue stack. And when evaluating a provider, you need to understand whether it should serve as a system of record and, if not, how it integrates with existing systems while preserving their integrity.

This systems-of-record layer forms the foundation of the revenue stack.

Above it sits the workflow automation and agentic layer. Before the rise of agentic AI, workflow automation platforms helped synchronize data and orchestrate processes across systems of record and the broader revenue stack. Representative vendors include Zapier and newer players such as n8n.

AI agents represent the next evolution of workflow automation. Rather than following predefined rules, they can operate autonomously toward a goal. They can also replace some of the workflows embedded within the revenue applications that make up your stack.

Agents can be specialized or custom-built. While many practitioners are experimenting with building their own, most organizations should assume a mix of vendor-provided and custom agents.

Agents are now coming from all corners of the market:

• Foundation model and AI platform providers such as Anthropic, OpenAI, and cloud hyperscalers AWS, Google, and Microsoft.

• Workflow automation providers such as Make, n8n, and Zapier have added agent-building capabilities.

• Your systems of record providers have likely embedded agents directly into their applications.

• Large enterprise application providers such as HubSpot, Microsoft, Oracle, Salesforce, SAP, and ServiceNow are increasingly offering agentic platforms that extend beyond their software and can orchestrate work across a broad range of business and sales processes.

• Ultimately, every application across the revenue stack will embed agents.

• New agent-native platforms purpose-built for revenue teams, such as Lindy AI, Lyzr, Relevance AI, Rox, and Swan.

Before introducing new providers into your environment, start by inventorying the agentic capabilities already available from your existing strategic vendors.

While the debate over agent development strategy rages, from customizing prebuilt agents versus building from scratch to selecting the right platform, too little attention is being paid to the context that makes agents effective.

The middle box of the agentic layer is intended to rebalance that discussion and foster the design of the context cornerstone, assembling elements such as semantic definitions and connected datasets that enable effective agents. And when using the blueprint to map a vendor’s offering, elements such as context graphs and MCP should be captured there.

That’s all for this first double-click. Comments welcome!

Last week, Zendesk held its annual user conference, Relate, alongside an Analyst Day. The event marked the culmination of a broad set of organic and non-organic developments that have re-shaped Zendesk’s product footprint into a comprehensive service suite spanning both customer and employee service, with AI embedded natively across the platform rather than dependent on third-party capabilities.

To complete its suite and accelerate its roadmap, the company has become a serial acquirer, snapping up eight companies in the last three years: Cleverly (AI triage and assistance), Tymeshift (workforce management), Klaus (quality management), Ultimate (customer service automation), Local Measure (contact center), HyperArc (GenAI-powered analytics), Unleash (enterprise search), and Forethought (agentic AI).

At the event, Zendesk showcased its Resolution Platform, now organized around four core solutions, each serving as an entry point into the broader suite:

1. Customer Service: Zendesk’s historical core solution.

2. Employee Service: Formally launched as a standalone solution last year and now expanded with IT asset management (ITAM) capabilities.

3. Contact Center: Added through the Local Measure acquisition, the solution powered by Amazon Connect Customer is now fully integrated into the Zendesk platform. The offering includes Zendesk’s agent workspace and reporting capabilities, leveraging the HyperArc acquisition. Zendesk shared that it has added more than 100 new customers since the acquisition.

4. AI Agents: Built on the Forethought acquisition, the solution remains available for non-Zendesk customer service and contact center environments while being integrated into the Zendesk Resolution Platform.

Relate’s theme was the autonomous service workforce, underpinned by a vision of AI and human agents working together, with AI agents spanning both pre-built, highly specialized agents and custom-built agents created through its newly introduced Agent Builder.

Deon Nicholas and Sami Ghoche, co-founders of Forethought, detailed one of its key technologies, the self-improving loop, which now underpins Zendesk’s new Resolution Learning Loop. The loop begins with AI analyzing historical data to surface knowledge gaps and identify automation opportunities, detecting patterns that can be translated into either knowledge base content or agent-driven workflows. It then evaluates conversations across multiple dimensions using customer feedback signals, structured topics, and QA performance metrics such as churn risk. Based on these insights, the system generates and refines agents, which are validated through simulation-based test suites and then deployed via controlled A/B testing in production. Today, these stages are AI-driven with human approval gates for reviewing recommendations, validating test outcomes, and approving production rollouts. The new AI Studio is positioned as the control layer that will progressively automate this loop end-to-end, enabling increasingly autonomous agent improvement.

Zendesk also provided interesting details around its data platform. Unlike many CRM platforms that require custom objects and workflows to accommodate customer-specific business processes, Zendesk has been able to preserve a consistent core data model for the ticket lifecycle and ticket-to-contact relationship. This provides a much more readable context layer for AI and agents.

CEO Tom Eggemeier has emphasized Zendesk’s AI ambitions, sharing progress through AI annual recurring revenue (ARR), which he stresses includes only revenue from AI products. The company reached $250M last quarter and now expects to exit the current fiscal year, ending January 2027, at $400M.

In the backdrop of traditional product category boundaries coming down and with so many vendor options available for AI, differentiation was a recurring theme in discussions. These conversations are often centered on AI capabilities and the data that feeds them. What stood out is that when asked why they are with Zendesk on their AI journey, customers consistently pointed to its people, both in terms of expertise and a genuine commitment to making them successful.

I attended the TP Immersive conference earlier in May.

With Jorge Amar recently appointed as CEO, it was a timely opportunity to take the pulse of an industry at a crossroads. In February, he took the helm of the company founded 50 years ago by industry legend Daniel Julien, succeeding him.

Jorge came from McKinsey, where he was already working on TP’s strategic transformation before taking the CEO role. He stepped in with a clear conviction: TP sits at the center of the workforce transformation AI is about to unleash.

BPOs are first in line for AI disruption. A significant portion of their revenue still comes from the “butts and seats” model built around entry-level roles that AI agents are poised to replace. Under mounting pressure, BPOs are reshaping their business profiles by diversifying into higher-value services, moving up the value chain through more end-to-end process ownership, and shifting toward outcome-based models.

TP operates a large workforce and has to handle significant turnover. Over time, that has made it highly proficient at hiring, onboarding, and continuous training. This creates a foundation for transitioning workers into the new roles that Agentic AI will inevitably create.

Interestingly, a growing number of businesses struggling to realize the expected financial returns from their AI initiatives are turning to the BPO giant to take over functions outright, including the people, with the charter to optimize them.

BPOs are also “project factories.” They manage many projects and, with slim margins, profitability depends on disciplined execution, particularly around timing and delivery. TP is no exception. It worked on more than 500 AI projects last quarter alone, building deep expertise across both the technical and change management dimensions.

Last year, TP introduced TP.ai FAB, a platform designed to support its vision of humans and AI working together.

TP.ai FAB is built on an open framework that integrates with clients’ existing environments and incorporates technologies from selected partners, including Ema, Kore.ai, Parloa, and Sanas.

At the event, TP shared how it built five agentic solutions on top of this framework:

• FAB Assist delivers a complete work environment for human agents, combining AI assistance, KPI-driven coaching, and Sanas voice experience improvements

• FAB Connect provides an agentic customer self-service solution with escalation to human agents

• FAB Operate optimizes traffic orchestration across contact centers and agentic self-service based on customer intents and desired outcomes

• FAB Collect for collections and FAB Growth for sales are end-to-end solutions designed to maximize revenue outcomes through AI.

These solutions position TP to take on more outcome-driven engagements. One challenge, however, is that in multi-provider enterprise environments, the technology stack is often dictated by the customer.

Jorge explained that this transformation requires a fundamentally different conversation with customers. Success depends on the ability to deliver meaningful insight into what businesses should consider. Over the years, TP has built deep expertise across industries and business processes that it now needs to fully harness to earn that strategic seat at the table.

TP’s position at the intersection of technology, people, and large-scale delivery expertise gives it the right to participate in the emerging service market disruption. Key to its success will be the ability to realize definitional outcome-based projects within its customer base.

On the heels of its 1Q26 earnings call, Twilio hosted its Analyst Summit and Signal conference, where it unveiled new products alongside a new positioning.

Before getting there, it is worth pausing on the company’s remarkable turnaround. Once a market darling, Twilio exited 2022 with over $1B in losses, $1B in debt, and gross margins below 50%, entering a period of significant restructuring and strategic reset. Early 2024, Khozema Shipchandler, previously the company’s CFO, succeeded founder Jeff Lawson as CEO.

The financial turnaround was largely achieved in 2024, but it came at the cost of six consecutive quarters of stagnant growth, which led markets to question the durability of the long-term opportunity. In early 2025, Twilio shared a double-digit growth ambition. The latest quarterly results, 20% year-over-year growth after five consecutive quarters of sustained double-digit expansion, were rewarded with a sharp increase in market valuation, setting the stage for what the company is now positioning as its next chapter.

The company has spent the past couple of years unifying its portfolio, including its SendGrid (email) and Segment (CDP) acquisitions, an effort that culminated at Signal with the unveiling of a new Console giving developers a unified entry point across its products.

The next act for the company is to become the infrastructure for the agentic era, enabling continuous customer engagement with humans and AI. At Signal, it launched its conversation layer, adding three new building blocks to Conversation Relay:

Conversation Orchestrator enables continuous conversations across channels with both human agents and AI agents. It enables seamless channel switching, multimodal interactions, AI-human handoffs, and automated outreach for follow-up or notification.

Conversation Memory adds persistent context across interactions. It aggregates customer traits (profiles), identifiers, observations (signals extracted from conversations, such as behavior, sentiment, and preferences that can be manipulated semantically), and interaction summaries.

While leveraging Segment technologies, Conversation Memory is CDP-agnostic and can also connect to any CDPs, CRMs, and data warehouses. For AI agents, Twilio can further enrich context with knowledge through its Enterprise Knowledge ingestion module.

Conversation Intelligence lets you take actions in real-time on live conversation streams. It features a framework that uses language operators. These operators, either from Twilio or custom-built, include surfacing contextual knowledge recommendations, delivering real-time guidance, enabling intelligent routing and next-best actions, monitoring script adherence, and detecting sentiment and conversational signals.

Conversation Intelligence differs from Conversational Intelligence, Twilio’s post-conversation analytics solution, which is now evolving into a new product, Conversation Insights, that will combine the analysis of structured and unstructured data.

The conversation layer also includes Agent Connect, which integrates these building blocks into agentic platforms while abstracting away conversational complexities such as streaming and turn-taking. It supports OpenAI, Microsoft Azure, AWS Bedrock, Anthropic, and LangChain/LangGraph. It’s open-sourced and can be adapted to other frameworks.

Twilio has refocused on APIs and building blocks. That shift has reignited momentum with ISVs, a business that had been hindered by Twilio’s earlier ambitions toward packaged applications. In January 2025, Twilio disclosed that ISVs represented roughly a quarter of revenue, with the segment consistently outgrowing the broader business since then.

Serving both enterprises and ISVs across organizations ranging from SMBs to the world’s largest companies creates inherent positioning challenges. Twilio addresses the diversity of needs across these segments by marketing a collection of modules. While it makes it easy for builders to pick what they need, it can take some stepping back to see where the platform is heading.

Twilio is reshaping itself into an infrastructure provider for customer engagement in the AI world. Looking across its moves, three pillars are taking shape, each rooted in foundational requirements for customer engagement in the agentic era. First, conversation management with its newly unveiled layer. Second context, with Segment getting not only modularized but also embedded in other building blocks. Third identity and trust, with Twilio extending its authentication capabilities through its Stytch acquisition to support AI agents and expanding into compliance and fraud detection.

On April 23, AWS acquired NLX, and for a company with a strong build DNA, any acquisition is intriguing. Over 20 years, the cloud hyperscaler has made only eight that I could identify.

So, who is NLX?

Founded by two brothers, Andrei and Vlad Papancea, in 2018, NLX was born out of Andrei’s experience building conversational applications and the recurring effort required to navigate technology cycles, iterate on experiences, and expand the use cases these applications can support.

This led to the creation of Canvas, NLX’s conversation builder, designed to be to conversational application design what Figma is to UI design. The platform was built to insulate application creators from the complexities of channels, communication infrastructures, and AI technologies.

The obvious question: don’t all conversational AI platforms, including Amazon Connect, already offer a no-code builder?

While almost all conversational AI platforms feature no-code development environments, most remain fairly low-level and require significant technical expertise. NLX created a platform for business users that minimizes dependencies on developers.

Its building blocks abstract application logic across channels, enabling teams to compose multi-modal experiences that guide users to the right modality at each stage of a workflow.

NLX Canvas enables teams to rapidly iterate and expand use cases, a critical success factor for deploying and scaling conversational experiences that today often depends on forward-deployed engineers.

NLX has remained focused on large enterprises with complex requirements, spanning customer experience and other enterprise use cases. Its customer roster includes Comcast, Red Bull, Saks Fifth Avenue, and Toyota.

NLX’s vision aligns closely with Amazon Connect’s mission to create ready-to-use solutions. It already extends AWS Lex Flow Builder and Amazon Connect Flows.

Now part of Amazon Connect Customer, NLX will focus on customer service use cases on the hyperscaler CX platform, while support for other platforms becomes customer-driven.

At this stage of conversational AI adoption, many enterprises are prioritizing customer self-service and selecting best-of-breed platforms. NLX enhances Amazon Connect’s ability to compete in these self-service modernization initiatives.

At its What’s Next with AWS event, the cloud hyperscaler made several announcements, including a particularly ambitious one: the expansion of Amazon Connect into a family of agentic business solutions, which I want to unpack.

You knew Amazon Connect as contact center software. Reset that view: it now represents a family of agentic solutions that package AWS cloud services into ready-to-use offerings.

This announcement carries real strategic weight. A few years ago, a rumor circulated that a team inside AWS was making the case for acquiring HubSpot, sparking debate about the company’s developer DNA and its ability to move up the stack into business applications.

Today, AWS is providing a clear answer to that question. It is not only building business applications, but also using agentic AI to deliver a new class of solutions.

The infrastructure giant has built a $1B business with Amazon Connect, its CCaaS solution. Using Connect as the umbrella brand for its solution portfolio signals a clear intent: replicate the same playbook that drove Amazon Connect’s success.

Colleen Aubrey, Senior Vice President of Applied AI Solutions, outlined the AWS approach: it leverages Amazon’s operational expertise, targets enduring business problems its parent could not solve with commercial software, and infuses talent from Amazon to bring that expertise into AWS. Proprietary “secret sauce” built from these in-house solutions is packaged into AI agents that function as teammates.

In addition to Amazon Connect Customer, the renamed CX/CCaaS solution, the portfolio includes three additional offerings.

Amazon Connect Talent is an AI-powered hiring solution built on Amazon’s experience of hiring 250,000 associates each year for peak season. It turns job requirements into structured, skills-based assessments and AI-led voice interviews that candidates can take on their own terms. It applies science-backed scoring and evaluation models to rank candidates and improve matching quality at scale.

Amazon Connect Health builds on Amazon Pharmacy and One Medical. The goal is straightforward: reduce administrative and documentation work from providers, a huge issue in care delivery. It uses agentic capabilities across the patient journey, starting with scheduling and intake to handle appointment booking, insurance checks, and reminders to reduce no-shows. During the visit, it transcribes patient conversations in real time, surfaces clinical context and coding suggestions, and generates draft clinical documentation for provider review. Every recommendation is explainable, with supporting rationale available to clinicians. Post-visit workflows, including follow-ups and medication adherence tracking, are also automated.

Amazon Connect Decisions is the agentic evolution of AWS Supply Chain, addressing supply chain issues for manufactured goods. It features two AI agents, one handling the demand side and the other the supply side, that proactively surface issues and generate options. A third agent assists human decision-making, including the option to incorporate best practices from top-performing planners. The system is powered by proprietary prediction models developed from Amazon’s experience managing hundreds of thousands of SKUs across its retail and logistics footprint.

Discussing his December prediction that we will soon live in a world with over one billion agents, CEO Matt Garman highlighted that agentic AI is starting to enable things that weren’t possible before. Two in particular: the ability to be given a desired outcome and run the end-to-end workflow toward its realization, and the ability to free up time for humans. The Amazon Connect portfolio is aiming at delivering on that vision.

What I find most promising in AWS’s approach is that, unconstrained by any application heritage, it can tackle end-to-end processes without boundaries, exactly the scope agents are built for. It will be fascinating to watch it mature and grow.

11 years ago, I created an Inside Sales Technology Landscape. While widely known as a map of categories and companies in the space, it was originally designed as a blueprint for architecting a software stack supporting inside sales.

The framework proved remarkably durable. I expanded it over the years to cover the full sales software landscape. It held up through a decade of innovation... up until agentic AI emerged.

The need for a reset became clear when I published its latest edition last November. So when Seth Marrs reached out to discuss assembling a blueprint to guide the design of a revenue stack, I seized the opportunity to leverage his insights to build it.

Data and AI now enable organizations to segment and target their market with laser precision, on a continuously updated basis:

• Define ideal customer profiles (ICP)

• Build a full, enriched total addressable market (TAM)

• Map buyers and buying committees to expand each target account with the relevant contacts

• Segment markets with far greater granularity

• Identify and aggregate signals that indicate emerging demand or timing

• Build scoring models combining fit, segment, and signals to prioritize accounts and contacts, and trigger the right sales and marketing plays

This data-driven market definition and prioritization foundation becomes the backbone of both sales and marketing, driving tighter alignment. It hinges on acquiring and enriching company, contact, and signal data, and combining it with your proprietary data.

AI is the catalyst for organizations to rethink their sales processes, shifting focus to end-to-end Jobs to Be Done (JTBD). We’ve become addicted to adding a new point solution for every sales engagement “play,” creating a patchwork of tools with poorly-connected mini-workflows. The result is constant context switching and friction at every handoff, ultimately impeding execution.

We can group the core JTBDs into the following:

• Enable sellers and buyers

• Build awareness & seed demand

• Prospect through personalized outreach

• Orchestrate deal pursuit through close

• Capture & convert inbound demand

• Execute high-velocity outbound calling & lead distribution

• Sell with and through partners

• Structure & execute deals

• Retain, renew, and expand customers

Not all of these apply to every organization, and relevance depends on your sales motions—inbound vs. outbound, targeted prospecting vs. high-velocity sales, partner-led models (sell-through vs. sell-with), account development, and product-led sales. These JTBDs can be mapped to the software categories that enable them.

The fragmentation of sales stacks, combined with persistent disconnects between marketing and sales systems, has driven a shift: from application-level reporting and analytics to a cross-stack instrumentation layer that captures engagement and sales activity and delivers a single source of truth for forecasting, pipeline, and performance management.

Agentic AI is inserting itself into the stack as a foundational layer that can execute workflows and act on them. It relies on a system of context that aggregates all the information required to execute its missions. It sits alongside legacy workflow automation and data orchestration technologies.

In this construct, systems of record sit at the base of the stack as the systems of truth for core data elements, providing context for AI.

It’s impossible to provide full color and detail in a single post. In the coming weeks, I will continue to unpack this new framework and share examples of how you can use it to map existing providers and your as-is and to-be stacks.