Philomath Problem Solver Strategist Thinker Tech & Engineering Thoughts and Musings Founder @TheBluebearAI & https://teapawt.net/t.co/ndoIyq5Lb8
Los Angeles
Joined April 2009
- Tweets356
- Following573
- Followers89
- Likes90
While reviewing Pix2Code's sales feedback, we found won deal value being treated as revenue.
For an agency, winning the work, getting paid and making money on delivery are different things.
An AI choosing which leads to chase needs that distinction too.
Our demo narration passed the missing-word check: 15 out of 15.
That still didn't tell me whether I'd want a client to hear it.
Working on Pix2Code keeps exposing this gap: checking that AI content is complete isn't the same as deciding it's ready to send.
1/3
I think agentic software engineering is doing more than shrinking teams. It’s collapsing several roles into one.
For a small company, one engineer could become its technical lead, combining parts of the developer, forward-deployed engineer and CTO roles.
2/3
That person could run a team of agents to build and maintain the company’s software.
With enough leverage, they could become a fractional CTO across several businesses—giving smaller companies access to technical leadership they couldn’t previously afford.
Anthropic is making long AI conversations faster. My BlueBear handoff tests have me asking a different question:
Will the next agent know why a client rejected an idea?
That's what I'd test before trusting it with client work. Still checking.
claude.com/resources/article…
I tested Pix2Code on our own X replies. It caught a BlueBear pitch that didn't belong, but some rewrites still repeated the same point.
For an agency using AI for client communication, that's the problem: a cleaner sentence can still waste the client's time.
A client says, 'Can we try this next month?' The transcript captures it. The CRM doesn't. Nobody owns it. Useful agency AI would catch that loose end, link to the call, and ask the account manager to confirm a next step. That's a workflow worth fixing.
If AI makes execution faster and cheaper, what still earns a retainer?
I think it's knowing the client's business well enough to say: here's what's working, here's what needs to change, and here's what we're doing next.
Then following through. That work has to be visible.
A useful idea from @utkarshships: separate DECIDED from PROPOSED in AI handoffs.
For agencies, I'd also keep who approved it and the source. A new session shouldn't turn a brainstorm into a client commitment.
teapawt.net/utkarshships/status/21…
1/6 I think we're about to see a significant shift in IT services, software development and agency businesses.
AI is making technical execution faster, cheaper and more accessible. Selling capacity, technical skills and billable hours will be harder to differentiate.
5/6 The firms that build those relationships and become embedded in how their customers operate will be in a stronger position to compete.
WPP says it will run Gradial as a managed service for marketing teams.
My take for small agencies: the agent is only part of what clients buy. Approvals, exceptions and someone accountable after launch are part of the service.
enterprisesolutions.wpp.com/…
Agency builders: when a client agent goes off track, can your team inspect the run?
30s from BlueBear's dev build: model selection, context/cache details and a selected earlier turn's activity.
What's missing from your debugging view?
Pix2Code edit + ElevenLabs voice.
🤖 Made with AI