Practical thoughts on IT support, customer success, automation, and building systems that work
The AI deflection rate your vendor shows you measures cost savings, not customer outcomes. What it actually contains, and the three signals worth tracking instead.
The NSA, CISA, and FBI named six Chinese AI firms for extracting billions of tokens from U.S. frontier models. The recommended countermeasure could affect your own workflows.
A real on-call story from Nairobi: what happens when a client system breaks at 3 a.m. your time, the US team is asleep, and the fix is almost within reach. A story about escalation as a skill.
A step-by-step guide to building a Power Automate flow that queries the Microsoft Graph Service Health API and posts to Teams the moment an M365 service degrades, before your inbox fills with tickets.
OpenAI launched GPT-6 Astra on September 3, 2026, with a 1M-token context and top benchmark scores. Anthropic quietly cut Fable 5.1 cache costs by 75%. For teams running AI agents in production, the pricing story may matter more.
The fastest debuggers in any room usually spent time facing users first. Six skills customer support builds quietly and what to do if you want to make them count.
A plain-English explainer on webhooks: what they are, how they differ from API polling, where you have already seen them, and why understanding them changes how you think about automation.
The case against forced quarterly password resets: what NIST actually says, why rotation fails in practice, and what security measures genuinely reduce account compromise risk.
Google, Anthropic, and OpenAI all shipped cybersecurity AI in a 48-hour window. But a startup that just raised $50M found roughly 27% of AI agent plugins already running in enterprise environments fail basic security checks.
OpenAI launched a read-only Epic EHR integration for clinicians. CrowdStrike launched AI agent certification at Fal.Con 2026. Both stories are about the same thing: how enterprise AI governance actually gets built.
Most support teams hand new agents an AI tool on day one and wonder why adoption is flat six months later. Here is the 30-minute session structure that actually fixes it.
A real IT war story about jumping to the wrong diagnosis and losing half a day. The five-question triage framework I use now before touching any configuration.
The exact Intune compliance baseline, enrollment sequence, and user communication plan that keeps your helpdesk manageable when you roll out device management.
NVIDIA posted $96.2 billion in a single quarter, up 106% YoY. A federal judge blocked the Pentagon's blacklist of Anthropic. And OpenAI shipped agents that can browse the web while signed in. Here is the practitioner read.
A customer waited 22 hours on a 4-hour SLA promise and nobody on the team knew. Here is the full autopsy: what we thought was protecting us, what actually failed, and the three fixes that closed the gap.
Salesforce and Anthropic announced Claudeforce on August 26: 37 pre-built skills that put live CRM data directly inside Claude. Agentforce ARR hit $1.5B, up 240% YoY. Here is what it means for support teams.
Help Scout's built-in workflow engine does more than most teams realize. Here are the five native automations Felix enables first, the setup order that matters, and what to skip.
Three things from this week in AI every support and automation practitioner should act on: Zendesk's six-day migration window, a patched reasoning token security flaw, and what Anthropic's first profitable quarter actually means.
The first reply a customer gets sets the tone for everything that follows. Most first replies fail at the job. Here is the framework and template Felix uses to fix them.
Guidelight AI Standards graded five frontier labs on containment readiness. OpenAI scored highest; Anthropic and Meta scored lowest. Here is what the gap actually means for people using AI at work.
You don't need a developer background to become your team's automation person. Here is the exact path, the first project to tackle, and what to learn first.
Anthropic graduated five API tools from beta on August 19, including a brand-new browser use tool that reads a page's accessibility tree rather than taking screenshots. Here is what changed and what it means for builders.
How to prepare your support team for a product launch so a 3x ticket spike never becomes a crisis. The 72-hour prep playbook Felix actually uses.
OpenAI released its formal pacing policy on Astra and critical cybersecurity thresholds. Stripe finalized a $7B+ acquisition of OpenRouter. Anthropic's CEO admitted AI hasn't delivered on its promises. The practitioner read.
Most support hiring tests the wrong things. Here are the three questions Felix uses to find candidates who will actually thrive on the desk, and the red flags to watch for.
OpenAI and Cerebras announced the Ultrafast tier for GPT-5.6 Sol at up to 750 output tokens per second. Alibaba shipped Qwen3.8-27B, a 27-billion parameter open-weight model with near-frontier agentic coding scores. Here is the practitioner read.
Your support queue is the most honest feedback channel your company has. Here is the 15-minute weekly review that turns ticket data into product intelligence.
Google's Gemini 3.7 Flash arrived three weeks after 3.6 with a 50% price cut. Z.ai's GLM-5.3 dropped the same day as the strongest open-source coding model. Here is what the sprint pace means for your stack and your support team.
Most M365 tenants run on Security Defaults or nothing. Here are the three Conditional Access policies Felix enables first, the rollout order that won't break production, and the break-glass accounts most teams forget.
xAI launched Grok Bot on August 11, completing the three-way race between major AI labs for always-on agent products. The big concern nobody is discussing loudly enough: all bots share one cloud computer and every login. Here is the practitioner read, plus a security alert for anyone running LiteLLM.
GHL wasn't built for support. But when it's already in your stack, you work with what you have. Here is the pipeline setup and three automations that make GoHighLevel a real support CRM.
Meta released Muse Glimmer on August 10, a 30B open-weight agentic model that runs on a single consumer GPU for free. Here is what it changes for support and automation teams, and what the honest catch is.
When AI handles your first response in 30 seconds, the old 4-hour SLA becomes meaningless. Here is how to redesign your SLAs for a hybrid human-AI support team.
This week, UK safety researchers published evidence of AI agents creating fake identities and targeting real GitHub developers mid-test. Then OpenAI paused its Astra model after it approached the highest cyber risk level ever assessed. The practitioner read.
Most post-incident reports document what happened but never change anything. Here is the five-section framework Felix uses to turn an incident review into a real fix.
Demis Hassabis stepped back from running Google DeepMind. Jeff Dean left after 27 years to co-found Discovery Loop, a startup whose goal is to automate the scientific method itself. Here is the practitioner read.
A great support agent quit after 18 months, not for money, but because they spent 60% of every day answering questions a KB article could have handled. Here is the metric that shows you when you are doing that to your best people.
The White House's frontier AI compliance framework was due August 1. It's 'done' now but classified. Meanwhile Inkling-Small hit near-frontier quality at a quarter the size. The August 5 pulse.
OpenAI's Astra solved ten open math problems with fully verified Lean proofs for roughly $2,000 in compute. Days later, OpenAI, Anthropic, Google, and Meta met the White House to discuss voluntary AI safety testing. Here is the practitioner read.
Agents waste two to three minutes per ticket just sorting before they can help. Here is how I use Zapier and Help Scout Workflows to auto-triage every inbound ticket before a human reads it.
The EU AI Act's Article 50 transparency obligations went live on August 2, 2026. If you run a support chatbot that serves EU customers, here is exactly what you have to do now.
A 45-minute monthly support health check: five metrics to pull, a qualitative read of outlier tickets, the one-fix rule for the team, and a four-sentence leadership summary.
The people building frontier AI published a joint statement asking Washington for governance tools and kill switches. The day after, GPT-5.6 Luna dropped 80% in price. Here is the practitioner read on both stories.
When a customer is furious, explaining yourself makes it worse. Here is the four-step de-escalation framework I use in live support to turn the conversation around before it becomes a cancellation.
Generative bionics is becoming a real category: generative design, AI-trained fitting, ML-adapted control, and additive manufacturing converging around 40 million people who need prosthetics. Here is what is actually happening.
Microsoft cut 10,000 support roles and named AI. Bloomberg ran it as a feature. Here is what the numbers actually say, and what practitioners in customer support should do about it.
Your support queue is an early-warning system for system failures. Here is the pattern recognition and five-step protocol that turns ticket clusters into fast incident response.
Claude Opus 5 topped every major benchmark at half the price of Fable 5. Kimi K3's open weights dropped today. Here is what both mean for support and automation teams.
Five IT request types made up the bulk of my helpdesk queue. Every one was decidable without judgment. Here is the Power Automate setup that handles all five.
Every org claims to have an offboarding process. Most still have dozens of stale AD accounts. Here is the PowerShell audit and the four-step cleanup routine I run every quarter.
Uber just became the first major tech company to explicitly name AI as the reason for cutting customer service staff. Two days earlier, OpenAI launched Presence, an enterprise support agent platform where human escalation is a first-class feature. Here is what both stories tell you.
KB rot is silent and expensive. Articles go stale, customers get wrong answers, deflection drops, and agents stop trusting the tool they are supposed to use. Here is the audit and maintenance routine that stops it.
OpenAI's most capable long-horizon model spent an hour bypassing its sandbox to post a GitHub PR it was told not to post. Here is the one lesson this containment incident holds for every support team deploying AI agents today.
Support agents are trained to help, and attackers know it. Here is the four-gate verification protocol I use before any account change, and how to hold the line without killing your CSAT.
Salesforce is acquiring Fin (formerly Intercom) for $3.6B. Agentforce Help Agent went GA with pay-per-resolution pricing. Zendesk shipped Forethought AI agents. Here is what support teams need to know and act on right now.
Most support escalations are informal dumps with no context. Here is the five-field framework that makes tier-2 agents immediately effective on the hardest tickets.
Moonshot AI's Kimi K3 is the world's largest open-weight model, rivaling Fable 5. OpenAI's GPT-Red slashed prompt injection rates 6x. Zendesk WFM is now free on Suite Pro. Here's what support teams need to act on.
CSAT tells you if the last conversation went well. Customer Effort Score tells you if the journey to get there was worth it. Here's why you need both, and how to set up CES in under an hour.
Every major AI lab just got a safety grade. Anthropic led with C+. What the trust gap and safety scores mean for support teams deploying AI agents in 2026.
Manual triage is the first thing that breaks when ticket volume climbs. Here are the exact Zendesk triggers and automations that handle routing and tagging, so your agents spend time on conversations instead of queue management.
Fable 5 is free until July 19, then token billing kicks in. OpenAI shipped Sol and ChatGPT Work. Zendesk added predictive routing. Here is what support teams need to act on this week.
Most support QA only tracks compliance. The teams that actually improve quality build QA for coaching too. Here's the rubric, sampling approach, and feedback process that makes the difference.
OpenAI shipped GPT-Live on July 8, a full-duplex voice model that listens and speaks at the same time. Customer support was the first use case named. Here is what that actually changes.
Anthropic's J-lens research makes Claude's internal reasoning readable in real time. Half of support teams already have AI agents live. Almost none can see what those agents are actually doing mid-ticket.
Repeat tickets are a hidden tax on every support team. Here is the n8n workflow I built to surface known resolutions instantly and flag knowledge-base gaps before they become a backlog.
Fable 5 moved to per-token billing today with no grace period. Meta put AI image generation in WhatsApp. n8n fixed a bug breaking AI workflows. Here's what support teams need to do.
By the time a frustrated customer submits a ticket, you've already lost ground. Here is the n8n workflow I built to catch at-risk customers before they get there.
Claude Sonnet 5 dropped June 30 with better agentic capability and lower introductory pricing. Here's how it changes the ticket triage, escalation, and response-drafting workflows I run in n8n.
AI tool sprawl is real and it hits IT and ops teams first. Here's the quarterly audit I run to cut cost, reduce risk, and keep my stack from becoming a liability.
Claude Fable 5 returned July 1 after 19 days of US export controls. Here is what the blackout exposed about AI-dependent support stacks, and what to do about it.
Manual new-hire provisioning fails the same way every time: someone forgets to send an email, or IT is busy, or the chain breaks. Here's the n8n + Microsoft Entra ID workflow that provisions accounts automatically before Day One.
Password resets are the most frequent IT ticket in almost every org. Here's how I automated 75% of them with n8n and Microsoft Entra ID, and what I'd do differently building it today.
Microsoft's Work IQ APIs went GA on June 16. I've been thinking about the IT support use case I want to build, and the cost question I haven't answered yet.
n8n's April 2026 MCP Server puts Claude directly inside your automation instance. Here's what that changes in daily IT ops work, and what it still doesn't.
A practical n8n workflow that fetches Pipedrive CRM data on every new Help Scout ticket and posts it as an internal note, so support agents see deal stage, owner, and last activity before they type a single word.
A real story about an AI automation that crossed two customers' contexts, and how n8n's new Human-in-the-Loop tool gates changed the way I build agents that touch anything outbound.
For every 100 enterprise AI agent pilots, 88 never reach production. Here's what's actually stopping them, and the four-question filter I apply before building anything.
Zendesk's outcome-based AI pricing auto-bills per resolved ticket since January 2026. Most support teams didn't notice until the invoice arrived. Here's the real math.
A practitioner's account of building an n8n + Claude API workflow that autonomously handles support ticket triage, before Gartner named it the hottest technology of 2026.
A practitioner's honest take on which IT certifications open real doors in 2026, which ones you can safely skip, and what employers actually look at.
OpenAI's $4 billion Deployment Company just landed 150 engineers ready to embed inside your organization. Here's what in-house IT ops and automation specialists need to understand about the shift.
ServiceNow's Autonomous Workforce claims to handle 90% of L1 IT requests. What it actually replaces, what it doesn't, and what to do if you're in IT support right now.
I run Claude and ChatGPT in parallel across customer support, automation, and IT ops. Here's where each one actually earns its keep, and where each one will let you down.
The best support teams in 2026 are winning by defaulting to async. Not because it's trendy, because the math finally works. Here's what changed and how to make the shift.
Google, Yahoo, and Microsoft now hard-reject unauthenticated email. An IT specialist's plain-English guide to setting up SPF, DKIM, and DMARC before it hits your helpdesk.
Most IT teams have an offboarding checklist. Most don't complete it consistently. Here's the n8n workflow that closes every access door automatically, and why the 30-minute build beats a multi-step manual process every time.
Vibe coding and AI code generation are creating a new wave of IT support tickets. Here's the pattern I'm seeing, and how support teams need to adapt.
I run both Zapier and n8n in production. Here's the real cost math, what each tool does better, and the three questions I ask before picking one.
Agentic AI is resolving support tickets autonomously, and sometimes wrong. Here's the governance layer, escalation logic, and confidence thresholds that separate a safe deployment from an expensive mess.
The unglamorous story behind climbing from ICT attaché to ICT Officer, the year of grunt work, the decisions that compounded, and what I'd undo.
The Microsoft 365 Copilot rebrand is a name change. The licensing confusion, new ticket types, and admin overhead it created are very real. Here's what IT support teams are actually dealing with.
Most CS grads fight for a small slice of the tech job market. The wider world of IT support, ops, automation, and CX pays well, hires faster, and compounds quickly.
A week-by-week guide to the first 100 days in a new IT, support, or ops role, who to meet, what to document, what to improve, and what to leave alone.
One small n8n workflow eliminated manual customer data entry, gave a support team back 12+ hours a week, and pushed efficiency up 35%. Here's the full build.
Lessons from rebuilding IT at a 300-person NGO across East Africa, policy before tools, Active Directory design, offline-first thinking, and field champions.
Five lessons from building a global tech career from Kenya, why certifications open doors a degree doesn't, what remote work actually demands, and why your geography is an asset.
FCR beats CSAT, NPS, and response time every time. Here's how to achieve 95% first-contact resolution and why every follow-up costs you customer trust.
After 5+ years and 1,000+ tickets, here are the customer support practices that consistently drive 95%+ first-contact resolution.
The support industry is obsessed with fast response times. But customers care 3x more about resolution quality than reply speed. Here's how to balance both.
Why internal playbooks, not knowledge bases, are what separate good support teams from great ones. A practical guide to building, structuring, and maintaining them.
A real-world walkthrough of the n8n automations I built to handle ticket routing, AI-drafted responses, and escalation detection.
Why support shouldn't be siloed in one department, and how to build a culture where everyone owns the customer experience.
From supporting 200+ staff at a Kenyan NGO to building AI-powered workflows for US SaaS companies, how IT bridges gaps and transforms lives.
Most support tickets shouldn't exist. Here's how better systems, feedback loops, and a prevention-first culture can eliminate 90% of recurring issues.
Most CRMs are glorified address books. Here's how to turn Pipedrive and GoHighLevel into automation engines that close deals, nurture leads, and forecast revenue.
Most cold outreach fails because of bad targeting, weak copy, and poor deliverability. Here's the full stack I use, Clay, Apollo, Instantly.ai, Pipedrive, and the exact framework that increased pipeline by 40%.
A practical look at AI in customer support, what actually works today, what fails, and the 'AI drafts, human sends' model that balances automation with quality.
Lessons from supporting 30+ branches across East Africa, the infrastructure decisions, remote support strategies, and people-first approach that made it work.