Helping support teams shift to a new agentic way of working.
San Francisco
Joined February 2023
- Tweets1.5K
- Following10
- Followers1.8K
- Likes640
Last week, we shipped 84 updates.
This includes AI-powered field generation, deeper API reliability, and continued agent quality improvements.
// PRODUCT PLATFORM
▸ Public API responses now include stable, machine-readable error codes.
▸ Issue fields can now be organized into folders and groups.
▸ Contact timezone is now automatically populated from Slack.
// AI PLATFORM
▸ AI agent search quality improved with automatic filter inference from natural language queries.
▸ Background agents can now complete more complex tasks without artificial tool-use limits.
// SUPPORT AGENT
▸ You can now give thumbs up and down feedback directly in the issue log.
▸ Backtest results now display escalation handoff reasoning.
// ACCOUNT INTELLIGENCE
▸ AI-powered formula generation is now available for account and contact fields.
▸ Added a milestone completion date field that auto-sets and can be manually edited.
// SUPPORT SYSTEM
▸ You can now filter issues by last customer message and last agent message separately.
▸ Phone trigger templates are now available.
And 73 more updates, improvements, and fixes behind the scenes.
Full changelog: usepylon.com/changelog
Last week, we shipped 118 updates.
This includes expanded semantic search for AI agents, a new pull-based team assignment model, and continued platform API growth.
// AI PLATFORM
▸ AI agents can now search issues, call transcripts, articles, notes, and survey responses semantically.
▸ Background agents can now be triggered on a recurring schedule.
▸ Agents can now view workspace roles and permissions, and update support hours and holiday schedules.
▸ Exposed APIs to create and update AI skills through Pylon MCP.
// SUPPORT AGENT
▸ Agent guidance character limit raised to 15,000 characters.
▸ Redesigned the backtesting UI for testing agent behavior.
// SUPPORT SYSTEM
▸ Pull-Based Team Assignment is now available in beta, letting reps claim work from a shared queue.
▸ Added a consolidated SMS feed showing message history across related SMS issues for a contact.
▸ You can now preview staged holiday replacement schedules in Support Hours.
// ACCOUNT INTELLIGENCE
▸ Customer portal users can now view a consolidated issue list across all the accounts they belong to.
▸ Completion percentages now display for milestones and tasks.
// PRODUCT PLATFORM
▸ Custom objects now support paginated grouping and grouping by user fields.
▸ Contributors can now submit KB article drafts for review directly from the issue-created article flow.
▸ Profile alias names now support slashes, enabling aliases with pronouns.
And 99 more updates, improvements, and fixes behind the scenes.
Full changelog: usepylon.com/changelog
We moved from manual support and Claude workflows to Agentic Support. Here's what changed for our own team:
➤ Median first response time: 72% faster
➤ 62% of tickets now have most of the work done by agents
➤ 64% more capacity per person
ordnl.link/OV85wlU
Last week, we shipped 113 updates.
This includes a major refresh to the agent chat experience, Health Score Builder reaching general availability, and continued platform API expansion.
// AI PLATFORM
▸ Major refresh of the agent chat experience, with a new session greeting, timeline-based thinking visualization, and improved page context.
▸ Added a new agent detail view.
▸ Added a test pane for validating agent behavior before deploying.
▸ You can now attach product issues from Linear, Jira, and other trackers directly to Pylon issues via MCP.
▸ Every agent now gets an agent-managed knowledge base by default.
// SUPPORT AGENT
▸ You can now simulate and view V4 escalations in agent Preview.
▸ Skills now support account-based eligibility filters, matching v3 runbook behavior.
// SUPPORT SYSTEM
▸ Sender email addresses are now always visible inline in email threads.
▸ You can now export route redirects to CSV.
▸ Clicking an unfurled link preview now opens the full work object with details.
// ACCOUNT INTELLIGENCE
▸ Health Score Builder is now generally available, with weighted components across Usage, Support, Engagement, Commercial, and Relationship Sentiment.
▸ Added a task import API to support bulk data migration.
// PRODUCT PLATFORM
▸ You can now hide agent names in customer portal responses, displaying your team name instead.
▸ Public API now supports creating and updating draft replies on issues.
▸ Added a Connected Members tab to the Slack integration settings.
// INFRA & SECURITY
▸ Added support for HIPAA healthcare EDI attachment file types.
And 97 more updates, improvements, and fixes behind the scenes.
Full changelog: lnkd.in/g5ajyui7
Last week, we shipped 115 updates.
This includes more capable agents, better visibility into handling time and SLAs, and configurable account health scores.
// AI PLATFORM
▸ Assist Agent can now set To, CC, and BCC recipients on issue drafts.
▸ Assist Agent and Background Agents can add comments to tasks.
▸ Cursor codebase integration is now available in issue Assist.
▸ Teams can configure Knowledge Base article access separately for agents and end users.
▸ Agents can find relevant untranslated articles when working in translated Knowledge Bases.
▸ Agent-authored messages are now visually distinct from teammate messages.
// SUPPORT SYSTEM
▸ Active Time is now live for Enterprise plans, with per-user handling time, live session timers, analytics, and API access.
▸ SLA breach detection is faster.
▸ Splitting an issue preserves its Slack connection on both resulting issues.
▸ The chat widget can serve Knowledge Base content in each visitor’s preferred language.
// ACCOUNT INTELLIGENCE
▸ Health Score Builder is now available in beta, with configurable signals and transparent component breakdowns.
▸ Customers can create tasks and share files directly in the customer portal.
▸ Project templates support task dependencies and relative due dates.
// PRODUCT PLATFORM
▸ Restricted issues support individual collaborators without expanding team-wide visibility.
▸ CSV exports now handle datasets with more than 200,000 accounts.
▸ The API supports creating Issue Groups and linking parent and child issues.
▸ The Pylon MCP connector supports custom task statuses.
// PRODUCT INTELLIGENCE
▸ Product areas can be generated automatically, with progress visible in the UI.
And 97 more updates, improvements, and fixes behind the scenes.
Full changelog: usepylon.com/changelog
Agentic Support is the new operating model for support teams.
This Wednesday (8/28) see how @evelegalai made the transition.
This will be a joint webinar with Michael Lauerman (Principal Technical Support Specialist @ Eve) and Advith (Co-Founder & CTO @ Pylon).
Register to join live or watch on demand:
ordnl.link/ShaczGc
Customer support is going agentic.
Today, we’re introducing the new Pylon, the first platform built for Agentic Customer Support. Available in beta now.
For decades, support teams have worked tickets one at a time. They gather context, investigate across systems, coordinate with other teams, decide what to do, and take every action by hand.
That way of working is changing.
➤ Before: humans did the work
➤ Now: humans direct agents to do the work
Most AI support products focus on deflecting simple questions before a human sees them. Agentic Support addresses the rest of the job: investigation, coordination, judgment, and action.
Pylon’s agents run on a context layer built from support, account, and product intelligence across every system your team uses.
Here’s how it works:
➤ Background Agents investigate tickets before anyone opens them
➤ Assist Agent lets humans review the evidence, ask follow-up questions, and direct work across connected systems
➤ Skills turn your team’s processes into instructions humans and agents can run
➤ Multiple agents work at once while your team steers the highest-value decisions
Humans stay in control. They inspect the work, correct it, and decide what happens next. As the team builds confidence, the same workflows can become more autonomous.
Cognition, Hex, Canvas Medical, Nominal, and other Pylon customers have been helping us test Agentic Support in early access.
Today, we’re opening it to more teams.
See how it works on July 20: bit.ly/4pkHrqG
Pylon retweeted
AI support produces nearly 2x as many unhappy customers.
Specifically, support tickets AI resolves without any human agents have the lowest percentage of promoters and highest rate of detractors.
Human-only resolutions have the most promoters, but AI + human resolutions are very close behind and surprisingly have slightly fewer detractors. This is measured on a CSAT rating.
To us this adds to the evidence that AI replacement workflows aren't actually the most productive for B2B teams. Instead we should keep investing in really great human-driven AI products.
And we expect AI-assisted support workflows will only get better over time.
—
This data is part of a larger article we partnered with Andreessen Horowitz on. Link to the full piece in comments.
Pylon retweeted
Our Anthropic bill is about to jump from $400K → $1.4M/yr.
Not because usage exploded, but because we're about to cross 150 seats.
Past 150 seats you're forced into Enterprise tier. Seats stop including any usage, every token bills at standard API rates. At our current run rate that's 3.5x overnight.
Unfiltered thoughts on AI spend:
1. We should spend tokens to grow as aggressively as possible. But most people (me included) aren't conscious of what they're spending.
2. Visibility comes first. People see their personal number and they're shocked. I accidentally spent $4,000 in 3 days in Claude Code.
3. For engineering the spend is clearly worth it. Pay for the best model, it saves more than it costs.
4. For a lot of other roles it's questionable. Apps nobody uses, skills someone already built. No ROI.
5. Spend limits are coming. We already require approval for more tokens on our support team.
The era of token-maxxing is coming to an end.
Pylon retweeted
The role of a support engineer is evolving.
It used to look like...
→ Long, manual issue investigations
→ Lots of back and forth with customers to get context
→ Ping an engineer for help with a ticket and wait
Now teams can...
→ Have AI run initial investigations
→ Ask AI to gather account context
→ Search the codebase directly with Claude Code or Cursor
The goals of support haven't changed (provide fast and high-quality help, make it easy for customers to reach you) but the tools are making it so much easier to achieve those outcomes now.
Next week we'll show you how we've been using AI tools for support, product, and GTM.
Robert and Nikolaj Pedersen from our support team will demo three AI workflows we actually use, including projects from our internal hackathon last month.
Join us June 10 @ 10 AM PT.
Link to register in the comments 👇️
Pylon retweeted
Product / design / engineering orgs of the future are vase-shaped.
Small product team
Big design team
Small engineering team
We've hit this point with AI where it's a significant multiplier for product and engineering capacity, i.e.:
→ It's easy for a lean engineering team to build and ship high-quality product super quickly
→ It's easy for individual PMs to do research and write PRDs at larger scale
For example, engineers can spin up a dozen agents and make more decisions much faster than before.
On the other hand design has yet to really hit its Claude Code/Cursor/Codex moment.
There are some AI tools that help with parts of the design process, but they just don't allow the same level of parallelization or quality right now.
So while many companies are hiring leaner engineering and product teams than we've seen historically, I think over time design teams will become more proportionally staffed.
Pylon retweeted
Now in Pylon...
Custom priority icons 🏷️
You can now replace our default priority labels and match Pylon's tags to your internal ranking system.
Enjoy!
Pylon retweeted
Our support team just crossed 100K tickets.
That's 17,355 across Jon Clark (leads the team and 1st support hire) and Matthew Nagy (2nd support hire).
32,000+ product questions, 26,000+ feature requests, and 23,000+ bug reports (top 3 categories).
5,000+ issues from our top 10 most active customers.
...all handled with a 93% CSAT.
On top of this, our support team is heavily involved in building and testing new Pylon products.
They're a huge part of the reason we can ship high quality features so quickly, and are always the first to pilot workflows we eventually productize and release to customers.
Congratulations to our support team for handling 100,000 (by now already 100,200+) issues. Thank you for keeping Pylon running 🚀
Pylon retweeted
New workflow: Send and set status ⚡
If you ever want to manually override AI auto-statuses in Pylon, you can now respond to an issue and set its custom status in the same action.
→ Draft your customer reply
→ Use the "Send" dropdown to set issue status on send
→ Use keyboard shortcuts to do it even faster
Make sure tickets land exactly where you want them.
Pylon retweeted
We got a new addition to the customer portal...
🍀 Feature requests 🍀
→ Turn on portal visibility for FRs
→ Select what metadata customers can see
→ Choose how FR statuses appear to customers
Now customers can stay updated on their feature requests, right alongside their open tickets and tasks.
Pylon retweeted
Here's how our first ever AI support hackathon went 👇️
What we learned from the event...
1/ For a lot of people, networking and just getting to chat with other attendees was super valuable in itself
2/ Our current customers got a ton of value out of meeting their Pylon CSMs in person (even if it was just a 15-30 min conversation in the middle of the day)
3/ It was awesome to see folks get excited about beta products we announced, but many people wanted deeper dives on use cases and how to set things up
We had a great time and got such good feedback that we're planning a similar series of customer-focused workshops for this summer.
Some of what we'll add based on your input...
→ Dedicated sessions with your Pylon CSMs
→ Event tracks so attendees can go deep on specific Pylon products or workflows they want to learn about
→ Small group sessions to facilitate different teams meeting each other and sharing their work
It's going be a very fun series so we can't wait to get more of the Pylon team and our customers involved.
Huge thank you to everyone who came out on May 21. It's still mind-boggling to us that people flew in from as far as the East Coast and the UK to be there!
See you again at the next one 🫡