Report as of 2026-10-29 PT · Source: Metix AI
Notes & method
01 Not a research race: research is 1.7% of headcount
Scope = current employees of 12 AI customer-service / CX companies, matched to each company's verified company page (same-name firms removed), excluding interns and advisors, investors and board members: 5,026 people, global, data as of 2026-04. All figures are aggregate statistics; the report shows no personal information.
By job title, research is 1.7% of headcount; adding ML engineers and “Member of Technical Staff” titles makes it 3.6%. Only 10 of 5,026 people have worked at a frontier lab such as OpenAI, Anthropic, DeepMind or xAI (13 on the broadest count) — and 4 of them now work in recruiting, 2 in AI data labeling; including Scale AI and Databricks brings it to about 1.0%. The largest research team belongs to the incumbent, Intercom (48 people).
Across the 10 AI-native companies, builders (engineering, product, design) are 40.4% and GTM (sales, support, consulting, marketing) 29.3%. New-style delivery titles such as agent engineer and forward-deployed make up 8.9%, classic pre-sales and implementation another 8.2% — each far above research (1.3%). Titles like Agent Engineer and Agent Strategist were coined by this wave; Forward Deployed Engineer is borrowed from Palantir.
At the AI-native companies, 16.3% have worked at Big Tech, 13.2% at legacy CX / SaaS software companies, 7.3% in consulting — and just 1.7% at a frontier lab or AI-infrastructure company. A few companies show a clear lineage: among Sierra's people, 14.5% previously worked at Google, 9.3% at Meta, 6.9% at Salesforce and 6.1% at Slack; Cognigy draws on telephony vendors such as Avaya and RingCentral. The exception is Crescendo: after acquiring outsourcer PartnerHero in 2024, 58.8% of its people have contact-center outsourcing backgrounds, versus 0.8% at the AI-native companies.
Sierra (54.9%) and Decagon (64.0%) are concentrated in the SF Bay Area, and most of their people joined in the past year (68.0% and 73.7%); about 1 in 10 works in recruiting. Cresta and Forethought, also US-based, have most people outside the Bay; Intercom has twin hubs in the US and Ireland; the rest are centered in Canada, Germany, the UK and Australia, while Crescendo's frontline teams sit in Honduras and the Philippines.
02Role mix: three kinds of company, three structures
The 12 companies only make sense split three ways: 10 AI-native agent companies, Intercom (legacy support software going AI), and Crescendo (a human-plus-AI hybrid that bought an outsourcer). All three have very little research; what differs is the balance of building versus servicing: customer service is 22.5% of Crescendo but 3.9% of the AI-native companies. Incumbents have another route — buying teams: NICE acquired Cognigy and Zendesk acquired Forethought; Intercom built its own, and its research team (48 people) is the largest of the 12.
| Metric | AI-native (10) | Intercom | Crescendo |
|---|---|---|---|
| People | 2,826 | 1,346 | 854 |
| Builders (eng, product, design) | 40.4% | 32.7% | 10.3% |
| GTM (sales, support, consulting, marketing) | 29.3% | 33.9% | 32.0% |
| of which customer service | 3.9% | 10.3% | 22.5% |
| Research | 1.3% | 3.6% | 0.0% |
| New delivery roles (agent engineer etc.) | 8.9% | 1.2% | 0.7% |
| Classic pre-sales & implementation | 8.2% | 3.9% | 1.1% |
03Talent sources: Big Tech and legacy software, not labs
Look at where people worked before: at the AI-native companies they come mainly from Big Tech and legacy software companies, almost never from labs. A few companies show a clear lineage that usually matches their founders and product; many others draw from everywhere, with no single pipeline. The first table lists each company's top four previous employers (share of that company's people); the second rolls them up by type of company.
| Company | Top previous employers | Lineage |
|---|---|---|
| Sierra | Google 14.5% · Meta 9.3% · Salesforce 6.9% · Slack 6.1% | Big Tech + the Salesforce family (Slack is Salesforce-owned) |
| Decagon | Meta 8.3% · Amazon 7.3% · Scale AI 6.7% · Google 6.3% | Big Tech + Scale AI |
| Cresta | Google 8.5% · Amazon 5.1% · Sprinklr 2.8% · Observe.AI 2.6% | mostly Big Tech, plus contact-center software peers |
| Cognigy | Avaya 5.3% · RingCentral 4.9% · Accenture 3.2% · Salesforce 2.8% | telephony & contact-center system vendors |
| PolyAI | Amazon 4.8% · Replicant 3.4% · Dialpad 3.4% · Google 2.7% | Big Tech + voice & communications companies |
| Crescendo | PartnerHero 39.1% · Concentrix 8.3% · Startek 7.8% · Allied Global 4.7% | contact-center outsourcing |
| Intercom | Meta 3.3% · Salesforce 3.3% · LinkedIn 3.2% · Accenture 2.7% | diffuse, no single pipeline |
| Parloa | Wayfair 4.0% · Salesforce 3.8% · Amazon 3.3% · HelloFresh 2.8% | diffuse, no single pipeline |
| Ada | Qualtrics 3.1% · IBM 2.8% · Clarabridge 2.5% · Amazon 1.9% | diffuse, no single pipeline |
| Forethought* | Cisco 4.2% · TrendKite 3.2% · Accenture 2.1% · McDonald's 2.1% | diffuse, no single pipeline |
| Previous-employer type | AI-native | Intercom | Crescendo |
|---|---|---|---|
| Big Tech (Google, Meta, Amazon, Apple, Microsoft, LinkedIn) | 16.3% | 13.7% | 1.8% |
| Legacy CX / SaaS software (Salesforce, Zendesk, Genesys, Sprinklr…) | 13.2% | 8.9% | 5.2% |
| Consulting (Accenture, McKinsey, Bain…) | 7.3% | 6.6% | 3.6% |
| Contact-center outsourcers (PartnerHero, Concentrix, Teleperformance…) | 0.8% | 0.4% | 58.8% |
| Frontier labs & AI infra (incl. Scale AI, Databricks) | 1.7% | 0.1% | 0.0% |
04A new job species: the agent engineer
This wave has created a new kind of job: someone who builds the agent, deploys it into the customer's systems and answers for the outcome. Sierra calls them Agent Engineers and Agent Strategists, Decagon Agent Product Managers and Agent Success Managers, and Cresta uses Forward Deployed Engineers and AI Agent Architects. The role is highly concentrated: 18.8% of Sierra and 16.3% of Decagon, while Ada, Cognigy and others have none and rely on classic pre-sales and implementation titles instead (Cognigy 15.8%, Ada 13.7%).
05Geography: only two are rooted in the Bay Area
Of the 12 companies, only Sierra and Decagon actually concentrate their people in the SF Bay Area. Cresta, headquartered in the Bay Area, has 68.2% of its people in the US and 16.2% in Canada, with just 18.7% in the Bay; Intercom has twin hubs in the US (35.7%) and Ireland (34.4%). The rest are centered abroad: Ada in Canada (66.0%), Parloa (64.7%) and Cognigy (59.9%) in Germany, PolyAI (51.5%) and Gradient Labs (81.1%) in the UK, Lorikeet in Australia (56.7%); Crescendo's frontline teams are in Honduras (39.9%) and the Philippines (26.5%). The Europeans are turning American fast: among people who joined in the past year, 56.0% of PolyAI's are in the US (versus 28.5% of earlier joiners), Parloa 48.4% (versus 13.9%) and Cognigy 39.7% (versus 19.0%, partly because it now belongs to NICE).
06Hiring momentum: most joined within a year
Counting when people joined the company (not their current role, so internal moves aren't counted as new hires), as of 2026-04, 73.7% of Decagon and 68.0% of Sierra joined in the previous 12 months, and the median tenure at both is just 7 months. For comparison, the figure is 26.4% at Intercom and 23.9% at Forethought.
07The 12-company panorama
All 12 companies in two tables. Figures are current employees visible in the Metix AI database (data as of 2026-04) — a sample, not official headcount: Sierra, for example, shows 537 people against about 721 on its company page. 42 interns and 127 advisors, investors and board members are excluded. Intercom is the incumbent, Crescendo the hybrid, the rest AI-native; † marks companies since acquired (Cognigy by NICE, Forethought by Zendesk). Ada's name overlaps with many unrelated organizations, so its count is a conservative floor; * marks fewer than 100 people, and Gradient Labs' research share in particular rests on a small sample.
| Company | People | Bay % | Median tenure (mo.) | Joined <12 mo. % |
|---|---|---|---|---|
| Intercom | 1,346 | 17.1 | 29 | 26.4 |
| Crescendo | 854 | 2.1 | 18 | 31.2 |
| Sierra | 537 | 54.9 | 7 | 68.0 |
| Cresta | 493 | 18.7 | 12 | 48.1 |
| Parloa | 399 | 1.8 | 16 | 40.2 |
| Ada | 321 | 1.2 | 38 | 26.9 |
| Decagon | 300 | 64.0 | 7 | 73.7 |
| PolyAI | 293 | 3.4 | 18 | 35.0 |
| Cognigy† | 284 | 0.4 | 19 | 28.1 |
| Forethought*† | 95 | 18.9 | 46 | 23.9 |
| Lorikeet* | 67 | 9.0 | 8 | 69.7 |
| Gradient Labs* | 37 | 0.0 | 9 | 51.4 |
| Company | Build % | GTM% | Research % | New delivery % |
|---|---|---|---|---|
| Intercom | 32.7 | 33.9 | 3.6 | 1.2 |
| Crescendo | 10.3 | 32.0 | 0.0 | 0.7 |
| Sierra | 37.4 | 26.8 | 1.9 | 18.8 |
| Cresta | 46.9 | 32.3 | 0.6 | 8.5 |
| Parloa | 32.1 | 27.3 | 0.5 | 6.3 |
| Ada | 34.6 | 34.6 | 0.9 | 0.0 |
| Decagon | 43.3 | 26.3 | 1.7 | 16.3 |
| PolyAI | 43.0 | 28.7 | 1.4 | 5.5 |
| Cognigy† | 39.8 | 31.3 | 0.0 | 0.0 |
| Forethought*† | 49.5 | 38.9 | 1.1 | 0.0 |
| Lorikeet* | 56.7 | 11.9 | 1.5 | 26.9 |
| Gradient Labs* | 48.6 | 18.9 | 18.9 | 0.0 |
Questions this report answers
- Which companies and how many people does this AI customer-service talent map cover?
- 12 AI customer-service / CX companies and 5,026 current employees: Intercom, Sierra, Decagon, Cresta, Forethought, Crescendo, Ada, Parloa, Cognigy, PolyAI, Lorikeet and Gradient Labs. Each company is matched by its verified company page, interns and advisors / investors / board members are excluded, and the data is as of 2026-04. Cognigy (acquired by NICE) and Forethought (acquired by Zendesk) are grouped by their AI-native origins.
- Is AI customer service a race for AI researchers?
- No. Research is 1.7% of headcount (3.6% including ML engineers), and only 10 of 5,026 people have worked at a frontier lab (about 1.0% including Scale AI and Databricks). At the 10 AI-native companies, builders are 40.4% and GTM 29.3%; new-style delivery titles such as agent engineer (8.9%) and classic pre-sales (8.2%) each far outnumber research (1.3%).
- Where does AI customer-service talent come from, and where is it based?
- At the AI-native companies, 16.3% have worked at Big Tech and 13.2% at legacy CX / SaaS software, versus 1.7% at frontier labs or AI infrastructure. Sierra's people come notably from Google (14.5%), Meta, Salesforce and Slack. Crescendo, which acquired outsourcer PartnerHero in 2024, is the exception: 58.8% of its people come from contact-center outsourcing. Only Sierra (54.9%) and Decagon (64.0%) are concentrated in the SF Bay Area.
- How should this report be cited?
- Metix AI Talent Intelligence, 2026-10-29. A delivery war, not a research race A talent map of the AI customer-service wave | Metix AI. https://metix.ai/reports/mapping/sf-ai-customer-service-talent-2026
Want the full talent x-ray of any company?
The company matching, role splits, talent sources and hiring momentum in this report were all produced with Metix AI. We can build the same for any target company or sector: full lists, role and source breakdowns, cross-company flow, and org structure.
Start a Mira trial