Jeff Joyce

Most B2B companies treat a form fill like the beginning of the buying journey. It usually is not. By the time someone fills out a form, they may have already: - Researched the category - Compared competitors - Read reviews - Talked to peers - Built a shortlist - Decided who they trust So if your sales and marketing motion only starts after the demo request, you are probably showing up late. That is the problem I break down in this video. The real issue is not always lead volume. It is timing.

I'm surprised this is still on TBH. lol

@claudeai This thread is useful if the examples include the annoying middle too: what the tool touched, what still needed a person, and which parts were reusable after the first run.

Claude Fable 5 has been out for a couple of days. Some projects people have already built with it:

@BillyJack_J Yep. The send button still needs a gate. I’d rather have the agent prepare the draft, sources, and proposed change, then make the human review step painfully obvious.

The AI ops work that matters for small businesses is not another chatbot in a side tab. It is the control plane behind it: what needs review, what an agent can do, what actually ran, and what proof came out. Agents can prepare the work. The send button still needs a gate.

@odedra_tarun That permission curve is real. I’d let an agent touch a calendar long before CRM fields, payment rails, or customer-facing copy — and each step needs a visible undo path.

AI agents are getting good at tasks. The limiting factor is trust. You'll give an agent your calendar before your email. Email before your CRM. CRM before your bank. The whole agentic future runs on a permission curve nobody's mapped yet.

@AnthropicAI This is a good AI rollout test because the work is real and messy, not just tutorial prompts. I’d want fellows leaving each nonprofit with the boring trail too: what Claude helped with, what still needed a person, and what got reused next week.

We’re launching Claude Corps, a national fellowship program matching people early in their careers with US nonprofits. We'll teach 1,000 people to use Claude, and pay them to use AI to advance their hosts’ missions. anthropic.com/claude-corps

@khemaridh Headless CRM is a fun name, but the real test is whether it remembers the awkward relationship context: last email, promised follow-up, who prefers text, and what should not get automated.

@ngxson Local/no-server is a fun constraint here. The thing I’d check next is whether the browser step can show the DOM state it acted on and where it stopped, not just that it clicked through once.

Hacking a quite fun demo: "agentic browser" via llama.cpp web UI The fun part: NO external server is required, all done via clever tampermonkey trick 😁😁

@MichLieben Memory gets practical fast here. If an agent is writing outbound or content, I want it remembering the positioning, killed drafts, and weird edge cases — not just the last prompt.

@defaiscope Yep. Per-step accuracy sounds great until the task has 40 browser moves and one weird permission prompt. I’d rather see a smaller loop with clear stop points than a fully autonomous run nobody can safely debug.

WILD: 🇺🇸 A browser agent that is 99% accurate per step still fails most of the time. Browserbase engineering lead Derek Meegan breaks down why fully autonomous is the wrong goal, and why security is actually the root of scalability.

@mattybme This is a good early AI job because the mess is already sitting in Salesforce. I’d want the report to separate “interesting anomaly” from “owner/action needed this week,” or it becomes another dashboard.

Asked an AI to review a Salesforce pipeline. Deals with no activity in 5,900 days. One rep carrying 90% of the forecast. Opps open since 2010. It was all there. Nobody had time to read it. crmdigest.io — 14 days free #Salesforce #RevOps

@BrainsAndTennis Vertical agents get interesting when the domain mess is part of the product. For GTM that means weird fields, stale owners, approval paths, and knowing when not to touch the row.

@robin_rrtx @chainlink Agent-readiness is a useful way to frame docs. I’d want examples that show the boring stuff too: auth, failed calls, stale state, and when the agent should stop and ask.

Career update: i've joined @chainlink's AI Enablement team to level up docs and dev ex for both humans and AI agents building with the Chainlink platform. the goal: to be at the forefront of agent-readiness. most devs already use AI in atleast some part of their workflow. Chainlink has long been the industry-standard oracle network powering defi and bringing capital markets onchain (price feeds, CCIP, VRF, compliance engines, etc. ) the world is figuring out how agents transact, verify, and act onchain. Chainlink is where that gets built. you are no longer building platforms for human developers, but also for agents. if an agent cannot easily build with your infrastructure, you've already lost half the battle. My job: make sure every developer — human or agent — can navigate that infrastructure without friction. excited to be here, and for what's ahead and ofcourse to be able to work with @pappas9999 @PriyankGupta03 @andrej_dev and many more talented folks there. LINK EVERYTHING.

@duncanriley @SiliconANGLE This is the kind of boring proof layer I’d want around agents. Not “trust me, it ran” — show the workflow step, what it touched, and the proof before the next system accepts it.

Diagrid brings cryptographic proof to AI agent and workflow execution siliconangle.com/2026/06/11/dia… via @SiliconANGLE

@linear I thought this was a rap or country music video for a moment.

Introducing coding sessions. Linear Agent can now triage issues, investigate the cause, write the fix, open a PR, and bring the code back for review. All shared with your team in Linear.

@ClaudeDevs Can we get another reset on that usage? ;) lol

@AgnoAgi Resume is one of those boring features that matters a lot. If a long-running agent drops mid-task, I want it to come back with the prior steps, failed calls, and next decision still visible.

@coinbase The guardrails line is the one I’d zoom in on. Once an agent has its own account and can pay for tools/data, I want a very boring limit log: what it was allowed to do, what it spent, and where it stopped.

Meet Coinbase for Agents. Give your agent its own account to: → Execute trades & manage your portfolio → Run autonomously under guardrails → Pay for data & research tools via x402 (coming next week) Agentic finance is here, and it's powered by Coinbase.

@perplexity_ai This is a better Computer test than a one-off browser trick. When research, code gen, connectors, and sandboxes all touch the same task, I’d want the run to show which sources/tools it used before I trust the final answer.

We're integrating Deep Research as a native skill inside Computer. It now connects to the agent harness that powers Computer, with access to search as code generation, long running sandboxes, connectors, tools, and licensed data. Available now to Pro and Max subscribers.

An agent can find the right company detail and still make a bad CRM update. Before it overwrites an account field, I want the compare view: - current value - proposed value - source link - when the source changed - which list, route, or report uses that field Otherwise the value can be accurate on its own and still break the follow-up/reporting layer.

@HubSpot @imkylecooke @drinkloverboy This is the AI-search check I’d want before a content sprint. Not just “are we visible,” but which prompt/source missed us and what changed after the page got fixed.

Props to @imkylecooke for this one. Also actual props. He was chaotic on set. Less chaotic when HubSpot AEO showed him @drinkloverboy wasn't showing up in AI search Try it free hubspot.com/aeo