@dmasin.bsky.social
🫵 Your turn!
What do you see as the biggest challenge in building AI agents for real-world customer support?
Reply below or DM us— I'd love to hear your thoughts 💭
What do you see as the biggest challenge in building AI agents for real-world customer support?
Reply below or DM us— I'd love to hear your thoughts 💭
January 7, 2025 at 11:55 AM
🫵 Your turn!
What do you see as the biggest challenge in building AI agents for real-world customer support?
Reply below or DM us— I'd love to hear your thoughts 💭
What do you see as the biggest challenge in building AI agents for real-world customer support?
Reply below or DM us— I'd love to hear your thoughts 💭
🌟 Our vision: SUPERHUMAN AI SUPPORT
AI agents that:
🧓 Grasp intent like seasoned humans.
💬 Handle multi-turn conversations effortlessly.
📖 Use historical context to *learn* and adapt.
It’s not just about answering—it's about delivering MAGIC ✨✨
AI agents that:
🧓 Grasp intent like seasoned humans.
💬 Handle multi-turn conversations effortlessly.
📖 Use historical context to *learn* and adapt.
It’s not just about answering—it's about delivering MAGIC ✨✨
January 7, 2025 at 11:55 AM
🌟 Our vision: SUPERHUMAN AI SUPPORT
AI agents that:
🧓 Grasp intent like seasoned humans.
💬 Handle multi-turn conversations effortlessly.
📖 Use historical context to *learn* and adapt.
It’s not just about answering—it's about delivering MAGIC ✨✨
AI agents that:
🧓 Grasp intent like seasoned humans.
💬 Handle multi-turn conversations effortlessly.
📖 Use historical context to *learn* and adapt.
It’s not just about answering—it's about delivering MAGIC ✨✨
🤖 So, how do we bridge the gap?
AI must move beyond "semantic similarity" and adopt reasoning approaches to handle:
1️⃣ Incomplete/misleading inputs.
2️⃣ Implicit, experiential knowledge.
3️⃣ Complex real-world scenarios.
We’re only scratching the surface. 🛠️
AI must move beyond "semantic similarity" and adopt reasoning approaches to handle:
1️⃣ Incomplete/misleading inputs.
2️⃣ Implicit, experiential knowledge.
3️⃣ Complex real-world scenarios.
We’re only scratching the surface. 🛠️
January 7, 2025 at 11:55 AM
🤖 So, how do we bridge the gap?
AI must move beyond "semantic similarity" and adopt reasoning approaches to handle:
1️⃣ Incomplete/misleading inputs.
2️⃣ Implicit, experiential knowledge.
3️⃣ Complex real-world scenarios.
We’re only scratching the surface. 🛠️
AI must move beyond "semantic similarity" and adopt reasoning approaches to handle:
1️⃣ Incomplete/misleading inputs.
2️⃣ Implicit, experiential knowledge.
3️⃣ Complex real-world scenarios.
We’re only scratching the surface. 🛠️
🔄 MULTI-TURN conversations = the real world
Support isn’t single Q&A. It’s multi-turn, flowing through phases:
1️⃣ Understand true intent.
2️⃣ Clarify missing details.
3️⃣ Provide answers & adapt.
RAG struggles here—it wasn’t designed for these dynamic, evolving exchanges.
Support isn’t single Q&A. It’s multi-turn, flowing through phases:
1️⃣ Understand true intent.
2️⃣ Clarify missing details.
3️⃣ Provide answers & adapt.
RAG struggles here—it wasn’t designed for these dynamic, evolving exchanges.
January 7, 2025 at 11:55 AM
🔄 MULTI-TURN conversations = the real world
Support isn’t single Q&A. It’s multi-turn, flowing through phases:
1️⃣ Understand true intent.
2️⃣ Clarify missing details.
3️⃣ Provide answers & adapt.
RAG struggles here—it wasn’t designed for these dynamic, evolving exchanges.
Support isn’t single Q&A. It’s multi-turn, flowing through phases:
1️⃣ Understand true intent.
2️⃣ Clarify missing details.
3️⃣ Provide answers & adapt.
RAG struggles here—it wasn’t designed for these dynamic, evolving exchanges.
🌀 Complex scenarios require reasoning
Imagine this: “I paid £1000 for a hotel, £300 was to be refunded. I added £100 for a restaurant bill but got only £150 back.”
A human sees the issue: the customer expected £200 but got £150. A RAG agent? Likely lost in irrelevant details.
Imagine this: “I paid £1000 for a hotel, £300 was to be refunded. I added £100 for a restaurant bill but got only £150 back.”
A human sees the issue: the customer expected £200 but got £150. A RAG agent? Likely lost in irrelevant details.
January 7, 2025 at 11:55 AM
🌀 Complex scenarios require reasoning
Imagine this: “I paid £1000 for a hotel, £300 was to be refunded. I added £100 for a restaurant bill but got only £150 back.”
A human sees the issue: the customer expected £200 but got £150. A RAG agent? Likely lost in irrelevant details.
Imagine this: “I paid £1000 for a hotel, £300 was to be refunded. I added £100 for a restaurant bill but got only £150 back.”
A human sees the issue: the customer expected £200 but got £150. A RAG agent? Likely lost in irrelevant details.
🕵️♀️ Human intuition is invaluable
Agents lean on years of experience, identifying patterns no knowledge base could fully document. E.g.: “Symptom X often leads to outcome Y”
RAG lacks this shortcut and struggles in troubleshooting. The knowledge base can’t keep up with EVERYTHING!
Agents lean on years of experience, identifying patterns no knowledge base could fully document. E.g.: “Symptom X often leads to outcome Y”
RAG lacks this shortcut and struggles in troubleshooting. The knowledge base can’t keep up with EVERYTHING!
January 7, 2025 at 11:55 AM
🕵️♀️ Human intuition is invaluable
Agents lean on years of experience, identifying patterns no knowledge base could fully document. E.g.: “Symptom X often leads to outcome Y”
RAG lacks this shortcut and struggles in troubleshooting. The knowledge base can’t keep up with EVERYTHING!
Agents lean on years of experience, identifying patterns no knowledge base could fully document. E.g.: “Symptom X often leads to outcome Y”
RAG lacks this shortcut and struggles in troubleshooting. The knowledge base can’t keep up with EVERYTHING!
🧠 Taking things @ face value = trouble
Customer might say “Card is broken” when the transaction was declined—or “Somebody took my $” when they forgot about a charge. Humans detect these subtleties. RAG systems? They respond LITERALLY, which can lead to irrelevant/harmful replies
Customer might say “Card is broken” when the transaction was declined—or “Somebody took my $” when they forgot about a charge. Humans detect these subtleties. RAG systems? They respond LITERALLY, which can lead to irrelevant/harmful replies
January 7, 2025 at 11:55 AM
🧠 Taking things @ face value = trouble
Customer might say “Card is broken” when the transaction was declined—or “Somebody took my $” when they forgot about a charge. Humans detect these subtleties. RAG systems? They respond LITERALLY, which can lead to irrelevant/harmful replies
Customer might say “Card is broken” when the transaction was declined—or “Somebody took my $” when they forgot about a charge. Humans detect these subtleties. RAG systems? They respond LITERALLY, which can lead to irrelevant/harmful replies
🤔 Understanding intent matters
Humans don’t just answer questions - they identify WHAT'S MISSING from a customer query. If user asks “Why can’t I pay?” human agent instinctively knows to ask for clarification. RAG agents? They rely on semantic similarity ALONE & miss the nuance
Humans don’t just answer questions - they identify WHAT'S MISSING from a customer query. If user asks “Why can’t I pay?” human agent instinctively knows to ask for clarification. RAG agents? They rely on semantic similarity ALONE & miss the nuance
January 7, 2025 at 11:55 AM
🤔 Understanding intent matters
Humans don’t just answer questions - they identify WHAT'S MISSING from a customer query. If user asks “Why can’t I pay?” human agent instinctively knows to ask for clarification. RAG agents? They rely on semantic similarity ALONE & miss the nuance
Humans don’t just answer questions - they identify WHAT'S MISSING from a customer query. If user asks “Why can’t I pay?” human agent instinctively knows to ask for clarification. RAG agents? They rely on semantic similarity ALONE & miss the nuance