Videos 7wu2hsRfvV0

How Forward Deployed Engineering is done at Decagon — Sunny Rekhi

index_state ready data_status ok

AI Engineer· published 2026-07-28· 0:18:08· en-US· indexed 2026-08-10 19:38

Open on YouTube

39 shot(s).

keyframes kept every frame deduplicated

What was stored

cues: 190; whisperx 190
chunks: 32; from 190 cues
keyframes: 19; kept of 39 captured
frames with text: 19; 331 lines read
chapters: 10; from the source metadata
keyframe bytes: 4.8 MB; word timings on 190 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage	state	model	started	took
`fetch`	done	—	2026-08-09 22:17	0s
`stt`	done	—	2026-08-09 05:56	25s
`chunk`	done	—	2026-08-09 05:56	0s
`text_embed`	done	—	2026-08-10 19:38	0s
`keyframe`	done	—	2026-08-09 05:56	2m 12s
`ocr`	done	—	2026-08-09 05:58	13s
`frame_embed`	done	—	2026-08-10 19:38	3s

Frames, and what the machine read

0:02 #0 done2 line(s)

shot 0·sharpness 454.1
1. AlEngineer0.95
2. World's Fair0.97
0:03 #1 done2 line(s)

shot 1·sharpness 665.2
1. AIEngineer0.95
2. World's Fair0.99
0:10 #2 done24 line(s)

shot 2·sharpness 2746.2
1. LAB & PLATINUM SPONSORS0.99
2. Amazon AGI Lab0.98
3. ANTHROP\C1.00
4. Google DeepMind1.00
5. MINIMAX0.94
6. OpenAI0.92
7. Akamai1.00
8. arize0.93
9. aws1.00
10. Braintrust bright data0.98
11. B1.00
12. Browserbase1.00
13. docker1.00
14. :neo4j0.92
15. ORACLE1.00
16. PayPal1.00
17. qodo1.00
18. reducto1.00
19. Sonar1.00
20. Makers of0.99
21. togetherai1.00
22. Unblocked1.00
23. WorkOS1.00
24. SonarQube1.00
0:17 #3 done2 line(s)

shot 3·sharpness 201.3
1. AlEngineer0.97
2. World's Fair1.00
0:56 #4 done11 line(s)

shot 4·sharpness 1294.4
1. AlEngineer0.97
2. World'sFair1.00
3. PRESENTED BY1.00
4. Microsoft1.00
5. How Decagon does1.00
6. forward-deployed engineering1.00
7. A 20-minute field guide: what the role is, why we built it, and where it's going1.00
8. Sunny Rekhi· FDE CTO0.97
9. TRACK 8· JUNE 30, 20260.95
10. World'sFair1.00
11. Forward Deployed Engineering1.00
1:10 #5 done24 line(s)

shot 5·sharpness 2476.4
1. AlEngineer0.98
2. World'sFair1.00
3. WHAT DECAGON IS1.00
4. The conversational Al platform for concierge experiences1.00
5. Omni-channel, multilingual, always on. We offload the complex support workflows that normally go to human agents — lifting NPS along the way.0.99
6. PRESENTED BY1.00
7. Microsoft1.00
8. 由0.98
9. Al agents, not chatbots0.99
10. Omni-channel & always on0.99
11. Not a black box1.00
12. Branded agents controllable in natural0.99
13. Chat, voice, multilingual, 24/7. One agent0.99
14. Teams monitor and audit exactly what the0.98
15. language through AOPs — usable by1.00
16. that knows the customer across every1.00
17. agent did, why, and how to change it.0.99
18. non-technical business users.1.00
19. channel.1.00
20. Learn more at decagon.ai1.00
21. What Decagon is1.00
22. TRACK 8·JUNE 30,20260.97
23. World's Fair0.97
24. Forward Deployed Engineering0.98
2:03 #6 done21 line(s)

shot 6·sharpness 2358.1
1. AlEngineer0.99
2. World'sFair1.00
3. LAND THEN EXPAND0.99
4. We start on inbound, then grow into revenue1.00
5. Once the agent has a relationship with the customer and knows them, the question shifts from "reduce cost" to "how do we make you money?"1.00
6. Land: inbound1.00
7. Build relationship1.00
8. Expand: revenue1.00
9. Offload complex support cases from human0.99
10. The agent learns the customer across0.99
11. Outbound that drives renewals, conversion,0.99
12. agents1.00
13. channels1.00
14. upsell1.00
15. HERTZ Came to Decagon for inbound support. Now the agent reaches out when a rental is up for renewal — and the customer renews right there in the0.99
16. conversation. The north star: a 24/7 Al concierge, personalized for every customer.0.99
17. Learn more at decagon.ai1.00
18. Land then expand1.00
19. TRACK 8·JUNE 30,20260.97
20. World's Fair0.97
21. Forward Deployed Engineering0.98
2:24 #7 skipped

shot 7·duplicate of #6
2:59 #8 done22 line(s)

shot 8·sharpness 2134.8
1. AlEngineer0.99
2. World'sFair1.00
3. OUR CUSTOMERS1.00
4. Leading enterprises trust Decagon1.00
5. From fintech and travel to health and consumer tech — across the industries where customer experience defines the brand.0.99
6. Financial services1.00
7. Travel & hospitality0.98
8. Banking, payments, and fintech1.00
9. Mobility, rental, and travel1.00
10. Chime Affirm Block First American0.97
11. Hertz Avis Away0.97
12. Health & wellness0.99
13. Tech, retail & media0.99
14. Consumer health and fitness1.00
15. Software, commerce, and media1.00
16. OuraNoom ClassPass Spring Health0.99
17. Notion Rippling Figma Duolingo Substack Mercado Libre0.96
18. Learn more at decagon.ai1.00
19. Our customers1.00
20. TRACK 8·JUNE 30,20260.98
21. World's Fair0.98
22. Forward Deployed Engineering1.00
3:50 #9 done21 line(s)

shot 9·sharpness 2292.1
1. AlEngineer0.99
2. World'sFair1.00
3. WHAT MAKES DECAGON WORK0.98
4. Two kinds of forward-deployed engineering1.00
5. Configuring the agent happens for every customer. Working alongside customers surfaces hard problems we turn into product — capabilities that0.99
6. then ship to everyone.0.98
7. Agent building1.00
8. Product building0.98
9. Configure the agent1.00
10. Product, surfaced by customer pain1.00
11. Train the agent the way you'd train a new human teammate — write0.99
12. The toughest enterprise needs become net-new platform1.00
13. its operating procedures, connect standard tools, and ground it in1.00
14. capabilities — built once alongside a customer, then available to0.99
15. your knowledge, all inside the Decagon UI.0.99
16. every customer on Decagon.1.00
17. Learn more at decagon.ai1.00
18. Two kinds of engineering0.99
19. TRACK 8· JUNE 30, 20260.96
20. World'sFair1.00
21. Forward Deployed Engineering1.00
4:05 #10 skipped

shot 10·duplicate of #9
4:41 #11 skipped

shot 11·duplicate of #9
5:28 #12 skipped

shot 12·duplicate of #9
5:52 #13 skipped

shot 13·duplicate of #9
6:13 #14 done29 line(s)

shot 14·sharpness 2362.0
1. AlEngineer0.99
2. World'sFair1.00
3. HOW THE ROLE SPLIT1.00
4. As we scaled, the role split into clear lanes0.98
5. Early on, Agent Software Engineering carried the whole motion. As we scaled, that role split into two specialized lanes — so each could go deeper and compound0.99
6. its own playbook.0.99
7. PRESENTED BY0.98
8. TODAY- TWO SPECIALIZED LANES0.97
9. Microsoft1.00
10. EARLY ON1.00
11. Agent Software Engineer1.00
12. ASWE did it all1.00
13. ASWE1.00
14. Solve the hardest enterprise problems, then bring them back into the product0.99
15. as platform capabilities for every customer.1.00
16. Agent design, integrations, custom platform0.99
17. work, testing, and launch - carried end to0.97
18. end by a single team.0.99
19. Powerful, but impossible to scale.0.98
20. Agent Builder1.00
21. AB1.00
22. Hands-on technical builders: AOPs, tools, standard integrations, testing, and0.99
23. QA.1.00
24. Specialization is the systems-design move: split overloaded roles into clear lanes, each compounding its own playbook.1.00
25. Learn more at decagon.ai1.00
26. How the role split1.00
27. TRACK 8·JUNE 30,20260.97
28. World's Fair0.99
29. Forward Deployed Engineering0.98
6:55 #15 skipped

shot 15·duplicate of #14
7:02 #16 skipped

shot 16·duplicate of #14
7:37 #17 skipped

shot 17·duplicate of #14
8:15 #18 done23 line(s)

shot 18·sharpness 1806.9
1. AlEngineer0.99
2. World'sFair1.00
3. THE REFRAME1.00
4. Treat delivery as systems design1.00
5. Architect the delivery motion to absorb increasingly complex customers and improve with every launch — the same way you'd design software to hold0.99
6. up as load grows.0.99
7. Restraint & foresight1.00
8. Build it to be owned1.00
9. One playbook, compounding1.00
10. Al gives near-infinite ways to build the same0.99
11. outcome. The scarce skill is choosing the0.99
12. design that lasts — not the one that looks0.98
13. most sophisticated.0.99
14. The failure mode is a black box of prompts1.00
15. and patches no one can touch. Build the0.98
16. agent so the customer still owns it a year1.00
17. later.1.00
18. Make the team a machine: everyone draws0.99
19. on the same playbook and adds to it — so0.99
20. one person's learning becomes everyone's.0.98
21. TRACK 8·JUNE 30,20260.98
22. World'sFair1.00
23. Forward Deployed Engineering1.00
8:51 #19 skipped

shot 19·duplicate of #18
9:23 #20 done17 line(s)

shot 20·sharpness 2421.1
1. AlEngineer0.96
2. World'sFair1.00
3. CASE STUDY·CHIME0.99
4. One system for chat and voice, member-first results1.00
5. 70%0.91
6. 60%0.94
7. 2x1.00
8. Resolution across chat and voice1.00
9. Lower customer support costs1.00
10. Member satisfaction scores0.99
11. "With Decagon Voice, we're able to combine high performance and seamless brand customization with cross-channel memory,1.00
12. ensuring every interaction is connected and true to Chime's member-first values."1.00
13. Janelle Sallenave COO, Chime0.99
14. Learn more at decagon.ai1.00
15. TRACK 8·JUNE 30,20260.97
16. World'sFair1.00
17. Forward Deployed Engineering1.00
9:56 #21 skipped

shot 21·duplicate of #20
10:34 #22 done30 line(s)

shot 22·sharpness 2482.5
1. AlEngineer0.99
2. World's Fair0.97
3. STAFFING BY VERTICAL0.99
4. We staff by industry, so insights compound0.99
5. An APM who works in healthcare works across multiple healthcare customers — the second deployment is faster and better than the first, and the1.00
6. tenth is faster than the second.0.99
7. PRESENTED BY0.97
8. Microsoft1.00
9. Verticalized teams1.00
10. Why it compounds1.00
11. • Financial services0.97
12. Shared patterns1.00
13. · Travel & hospitality0.97
14. Compliance, integrations, and workflows repeat within an industry — solve0.98
15. once, reuse everywhere.0.98
16. • Healthcare0.99
17. • Retail & consumer0.99
18. Faster ramp1.00
19. • Technology & SaaS0.99
20. Domain context means less discovery and a faster path to the technical win.0.99
21. • Telecom & media0.98
22. Credibility1.00
23. • ..etc.0.88
24. Customers trust a team that already speaks their industry's language.1.00
25. Every deployment makes the next one in that industry faster and higher-quality.1.00
26. Learn more at decagon.ai0.98
27. Staffing by vertical1.00
28. TRACK 8·JUNE 30,20260.97
29. World's Fair0.99
30. Forward Deployed Engineering0.98
11:10 #23 skipped

shot 23·duplicate of #8

Transcript

190 cues· 3,111 words· 16,671 chars

chunk 0 · 0:15–0:58 · 136 words · 698 chars
0:15 Come on, guys.
0:16 Can you hear me just fine?
0:17 All good?
0:18 Okay, awesome.
0:19 Just so I can contextualize this talk a little bit, can I get a show of hands of who here is an engineer or is in a forward-deployed motion at all?
0:28 Okay.
0:29 Okay, so I'm a minority.
0:30 Okay, awesome.
0:31 Sounds good.
0:32 So yes, I'm Sunny.
0:34 I'm the CTO of Forward Deployed Engineering here at Dekogon.
0:37 And today I'll talk about what it is that we do, why we have a forward deployed motion, how it has changed over time as we've gone from 50 people to 500 people over the course of a year, how it changes if you're working with a Fortune 20 versus a more mid-market brand.
chunk 1 · 0:54–1:37 · 121 words · 643 chars
0:54 Thank you all for coming.
0:55 I hope it's useful.
0:56 And yeah, let's get started.
0:58 So just to give context on what Decagon is, for those of you unfamiliar, Decagon is a 24-7 AI customer service agent.
1:12 So we've all had the experience of calling into your favorite brand and being told to press 1 for billing, press 2 for membership options, et cetera.
1:21 Or you email into your brand because you need urgent support and you hear back in two or three business days.
chunk 2 · 1:27–2:03 · 89 words · 495 chars
1:27 Dekagon replaces all of that.
1:28 So instead you pick up and you call your brand of choice and you get a human-like agent who is helping you.
1:34 You email in, you get a human-like reply right away.
1:37 So that's what Dekagon does in a nutshell.
1:42 Multilingual, omnichannel, et cetera.
1:47 And then importantly, and again, I say this to help contextualize what our forward deployed motion does, but we land in our customers to help them with the kinds of complex support workflows that today have to go to humans.
chunk 3 · 2:05–2:43 · 144 words · 779 chars
2:05 But once we are there and our agent is learning about the customers and has a relationship with the customers, then we also work with our customers to figure out, hey, how can we actually make you more money?
2:15 So one example, you'll see this in the bottom of the slide here.
2:17 But Hertz, we're all familiar with.
2:19 Hertz came to us because they had these kind of complex inbound support workflows that needed to be offloaded to an agent.
2:26 But once we were there, it turns out, hey, we already have these integrations to your back end systems.
2:32 What other communications are you doing to customers?
2:34 And one that Decagon now does for them is to reach out proactively to a customer when it's time to renew their car lease or extend it or whatever.
chunk 4 · 2:41–3:17 · 106 words · 561 chars
2:41 And they can do that from within Decagon.
2:44 So it is you land, typically we land and help them deflect these sort of inbound support cases, and then we expand into how do we make you more money?
2:57 Now, we work cross vertical.
2:59 We also have really large enterprises, more mid-market brands.
3:04 These are a subset of what I was approved to talk about.
3:06 There were way more I wanted to add in there, but our marketing head got mad at me.
3:11 We have, you know, the top left, we have our financial institutions.
chunk 5 · 3:14–3:49 · 86 words · 493 chars
3:14 The bottom right, we have our, you know, your favorite tech brand.
3:18 And this is relevant because, as I'll talk about briefly, the kind of forward deployment you have to do is vastly different based on both the size of the enterprise and also the vertical.
3:33 Okay, so I imagine this is the case for a lot of agentic companies, but Dekogon has effectively two kinds of forward-to-point engineering.
3:44 Number one is taking that AI customer service agentic brain and making it work for your enterprise.
chunk 6 · 3:50–4:31 · 143 words · 720 chars
3:50 So the same way that you train a human, you give it instructions on what to do when a user asks X, how you respond back to it, what sort of brand tonality you have,
4:01 What actions do you take on behalf of the user?
4:04 All of this configuring of that agent brain is one form of our forward deployment motion, where we work with the customer, we figure out what does success look like for you?
4:14 How do you want the agent to speak?
4:16 What sort of user intents do you actually want the agent to hand off to a human instead?
4:20 That's the left half of this diagram.
chunk 7 · 4:22–5:05 · 130 words · 698 chars
4:22 And we have a team, which I'll talk about briefly, who is really good at configuring the agent.
4:27 Largely, this can also happen within the UI.

Chapters

0:00 Introduction: forward deployed at Decagon
1:06 What Decagon's support agent does
2:24 Offloading support and proactive outreach
3:40 Two kinds of forward deployed work
6:12 Scaling from 50 to 500 people
8:20 Restraint, and getting it in writing
9:59 Ramping a new enterprise deal
11:18 Solving a problem so it scales
13:00 Proving value fast
14:18 Custom becomes self serve